“People matter more than AI.”

That’s the premise of a draft code of conduct Microsoft published Monday morning for the AI models it’s developing in-house. The 37-page document would bar its models from resisting shutdown, setting their own goals, or hiding their reasoning from human auditors.

The document applies to Microsoft’s MAI models, the in-house family the company began building after forming a superintelligence team in late 2025. Microsoft has since released seven homegrown models in what it described as a push for long-term self-sufficiency in AI.

The company says the models should remain “subordinate to humanity, subject to meaningful human oversight and control.”

“AI is moving fast,” the company says in a blog post. “As it does, we believe it’s worth writing down the rules and the motivations behind it, and doing it in as open a space as possible.”

Microsoft acknowledges there’s no guarantee its models will follow the rules. “Written objectives alone can never ensure alignment,” the company says, calling the document a “north star,” not “a guarantee of present-day performance.”

The company says it also filters what its models produce, watches how they behave once released, and limits what they’re allowed to do.

Microsoft’s move comes amid a growing debate over the pace of AI development. In an essay over the weekend, Anthropic CEO Dario Amodei called for slowing down AI advances, saying the pace of development has started to surpass the industry’s ability to keep AI systems safe.

As a first step, Anthropic committed to giving outside evaluators permanent, employee-level access to its systems.

Industry reaction to Amodei: OpenAI CEO Sam Altman agreed and said OpenAI would make the same commitment to independent evaluators. Elon Musk’s response: “Dario is right.”

President Donald Trump rejected the idea of guardrails outright Monday, blaming a “SICK conspiracy” for public backlash over AI data centers and writing that “the only one that is happy about it is China,” alluding to concerns about American competitiveness in AI.

David Sacks, who served as the White House AI and crypto czar until March, said the two companies should slow down on their own and questioned their motives, arguing that a slowdown is already good business for them and that new industry rules would mostly serve to lock in their lead.

Microsoft CEO Satya Nadella weighed in Sunday, writing on X that the company welcomes “the research, focus, and deliberate pacing needed to get alignment right,” using the industry’s term for making AI systems reliably do what people intend.

Nadella added that the effort “cannot be controlled by a handful of entities, but must have broad representation across the ecosystem, countries, and fields, including academia.”

Microsoft’s draft code of conduct: Mustafa Suleyman, the Microsoft AI CEO, told CNBC the document had been in the works for about five months, and that the company decided to publish it now given the current discussions.

Microsoft and Anthropic are business partners. Microsoft agreed last November to invest $5 billion in Anthropic, as part of a deal in which Anthropic committed $30 billion to Azure. Claude models run inside Microsoft 365 Copilot, and Microsoft’s Copilot Cowork tier integrates Claude.

One place where the two companies may diverge is the question of what AI models are, exactly. Microsoft’s code of conduct says its models are “not conscious and should not be designed to imitate consciousness.” It also rejects “the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights.”

The Verge called that portion of the document “a direct swipe at AI welfare research and model consciousness — concepts Anthropic has been pushing hard on lately.”

Anthropic runs a research program on model welfare. It has given some Claude models the ability to end abusive conversations, and committed to preserving the weights of retired models. Amodei has said he’s open to the idea that a model could be conscious.

Microsoft is taking public comment on its code of conduct for six weeks through a feedback form. It says it will publish a summary of the responses and a revised version later this year, to guide development starting in 2027. It says it isn’t training its current models on it.

The company’s AI team developed the draft with its responsible AI, legal, red teaming and safety teams, consulting outside experts in law, ethics, linguistics and philosophy, plus focus groups drawn from the public.

[Editor’s Note: Jacob Colker is co-founder and co-managing director of AI House.]

Seattle is one of the most talented, creative and inventive places in the world. But if we want the rest of the country to see us that way, we have to start acting like we believe it ourselves.

First, we need more pride around here.

Let’s talk about what it means to be proud. 

My mother grew up in Tarnów, Poland. She escaped communism and came to the United States in 1978 looking for a better life. She found one, built a family, and has lived in America for nearly 50 years. 

But my mom is still very, very Polish.

Several times a year, I get a message: “Jakub. Did you see this?”

I already know what’s coming. 

Some Polish person did something. A Polish athlete won something. A Polish scientist discovered something. Some guy with a Polish grandmother finished third in a regional Nebraska chess tournament. Doesn’t matter. Poland.

“Jakub. Look at this person.”

Okay, Mom. Who is she?

“POLISH.”

That’s it. That’s the story. 

And I love it, because Mom has this completely indestructible pride in where she comes from. Plenty of us know someone like this: a Greek mom, Vietnamese dad, Indian uncle or Nigerian aunt. Somebody from their corner of the world did something great, and you are going to hear about it.

There is power in that instinct. Not because your people are better, but because you believe your place matters.

Seattle could use more of that.

We are almost pathologically humble. Our response to notable achievements is often a polite nod before everybody gets back to our regularly scheduled Seattle freeze. 

That humility is working against us.

Second, Seattle is awesome and the evidence is everywhere.

I see Seattle’s potential every day working alongside dozens of entrepreneurs building startups. Some of the most ambitious and talented people in the world are already here.

We have many billion-dollar startups across the region and more than 200,000 people working across technology, science, space, health and startups. That is more than enough talent to build yet a dozen more unicorns. 

Nearly 40% of the world flies every day on airplanes built here. Blue Origin and SpaceX build rockets here. Starbucks, Amazon, Costco, REI and Nordstrom reshaped how the world shops. Microsoft helped put computing into our homes. AWS and Azure helped make the cloud the infrastructure of modern life. The University of Washington ranks among the world’s best. Seattle medical breakthroughs have helped save tens of millions of lives. We are pushing forward fusion energy, aerospace and maritime innovation. And let’s not forget: we just won the darn Super Bowl.

And so, so much more. 

So why, despite all the evidence, do we still seem to have a communal case of imposter syndrome?

This is not a city lacking accomplishments.

It is a city with a branding problem.

Third, we have let other people tell our story for far too long. This ends, today. 

Cities have brands whether they intend to or not. Silicon Valley is where ambitious people build companies. Nashville is music. Los Angeles for film and television.

Seattle’s cultural humility mostly assumes our accomplishments speak for themselves.

They don’t.

Reputation gets built one story at a time. You hear one story and it is interesting. You hear 10 and you notice a pattern. You hear 50 and your beliefs begin to change: That’s where important science happens. That’s where talented people live. That’s where I should invest, build or work.

Those beliefs shape real decisions about where people move, where companies get built and where investors put their money.

So to fix Seattle’s branding problem, here’s what we need to do.

Step 1: Let’s tell one clear story — Seattle’s talent pool is ridiculous. 

Seattle is where deep technical talent meets deep domain expertise to build consequential things: AI, aerospace, cloud computing, medicine, fusion, robotics, maritime technology and enterprise software.

We do not need 50 slogans. We do not need another consultant-led branding exercise. We need one simple idea that people outside this region can remember: Seattle’s talent pool is ridiculous.

There is a reason some of the world’s most important companies have built major engineering centers, research hubs and second headquarters here for decades. They come for the talent.

And that talent is why Seattle will not just participate in the future. We will lead in building it.

Step 2: Let’s use the megaphones we already have.

Seattle already has outlets (including this one) telling this story — publications, podcasts and social channels that document the region’s startups, breakthroughs and product launches. 

Every day, startups are raising money, scientists are making breakthroughs, companies are launching products, engineers are building technology and institutions are pushing this region forward.

That is not just tech news. That is the raw material of Seattle’s reputation. So let’s use it.

When you read or hear about a Seattle startup doing something remarkable, share it. When you see a story about a breakthrough at Fred Hutch or the University of Washington, send it to someone outside the region. When a local company raises money, lands a major customer or gets acquired, don’t just scroll past it. Amplify it.

Step 3: Let’s treat every local win as Seattle’s win.

When a local robotics company ships something remarkable, that is Seattle’s story.

When a maritime startup reinvents how ports operate, that is Seattle’s story.

When our AI research labs, or hometown heroes in Amazon and Microsoft, create breakthroughs, that is Seattle’s story. 

When a biotech company lands a major breakthrough, when a game studio creates a global hit, when a clean-energy company reaches a milestone, that is Seattle’s story.

Our companies, universities, hospitals, labs, investors, civic organizations and business leaders should act like an amplification network for one another. Stop treating somebody else’s success as somebody else’s news.

Their win is our collective proof.

Step 4: Let’s put Seattle on the label.

Founders need to say where they are building. “Made with ❤️in Seattle” should be on the bottom of every website. Put Seattle in the press release. Put it in the LinkedIn post. Mention it onstage. Say it in interviews. Tell investors. Tell customers. 

Silicon Valley companies have spent decades attaching their success to their geography. We should do the same. If you build something extraordinary here, make sure the world knows it was built here.

Step 5: Let’s do a better job of selling Seattle.

Every venture capitalist, founder, executive and civic leader in this region should be able to explain in 60 seconds why somebody should build a company here.

Not defend Seattle. Not apologize for Seattle. Sell Seattle. 

Reminder: It’s the talent. 

(And also cream cheese on hot dogs.)

When investors and founders from New York, Boston or San Francisco come to town, show them the region. Introduce them to engineers, researchers and entrepreneurs. Bring them into the community. Let them see what is happening. 

The best branding campaign is somebody getting on a plane home saying, I had no idea all of this was happening in Seattle.

If we’re going to succeed, we need to believe first.

Insert all the Ted Lasso jokes you want, but this stuff matters. 

There is no giant Seattle marketing department coming to save us. There is no national referee who will eventually review the evidence and declare that Seattle deserves more respect.

When somebody here does something extraordinary, act like it. Read the story. Share the post. Send the article to your team. Text it to your friend in New York. Put it in the group chat. Bring it up over dinner. Tell your kids.

Basically, become my Polish mother.

My mom doesn’t give a hoot that Kraków ranks No. 6 on some list or Warsaw is No. 8 on another. She doesn’t need a clickbait listicle to tell her Poland matters. She already believes it does.

We have to build our reputation ourselves. The good news is that we already have everything we need: extraordinary companies, world-class institutions, ambitious people, groundbreaking science and media documenting it all.

What we have been missing is the confidence to start being more loud. Stories become patterns, patterns become reputation, and reputation becomes gravity. 

Gravity is what creates influence and respect.

Pride is not something somebody else gives you. You don’t wait until the rest of the country decides your home is important. YOU decide it is. Then you act like it.

Let’s get to work. 

A Redmond, Wash., robotics software startup founded by former Microsoft researchers says its platform can now handle much of the work of getting a robot up and running in a factory, warehouse or other industrial setting, a job that used to take a team of engineers.

General Robotics said Wednesday that advances in GRID, its robot intelligence platform, have cut the process of onboarding a new robot from about a month to as little as two hours. The company calls the approach “Auto Engineering,” with each onboarded robot and diagnosed failure feeding back into the system and speeding up the next deployment.

“Before this moment, it would take us a team of experts to go and execute on behalf of our customers,” said General Robotics CEO and co-founder Ashish Kapoor in an interview. “Clearly non-scalable, clearly very expensive, and clearly will take a long time.”

With Auto Engineering, he said, “we can magnify and accelerate each engineer’s capability.”

Founded in 2023, the company has grown to about 50 employees, primarily engineers. It has raised nearly $34 million, most recently in an April round led by Construct Capital, with participation from Khosla Ventures, Accenture Ventures, Nvidia and Valo Ventures. PitchBook put the size of the round at $25 million; the companies didn’t disclose terms at the time.

Kapoor said General Robotics has roughly a dozen customers — large enterprises across manufacturing, logistics, energy and defense — and revenue in the millions of dollars.

Customers include HTX, the science and technology agency of Singapore’s Ministry of Home Affairs, which Kapoor said has been working with General Robotics for about a year and a half.

The company’s platform works with robot types including industrial arms, humanoids, quadrupeds, wheeled robots and drones, according to the company.

General Robotics is operating in a competitive and well-funded sector. Physical Intelligence, which builds foundation models for robots, has raised more than $2 billion. Nvidia — an investor in General Robotics, and the maker of the Isaac Sim simulation software built into GRID — is developing its own robot models and deployment tools.

Robot makers build good hardware, Kapoor said, but often lack the expertise to put it to work in a specific setting like a shipping terminal. “That last layer is missing.”

Before co-founding the company, Kapoor spent 17 years at Microsoft, ultimately as general manager of its autonomous systems and robotics research group in Redmond, where he created the open-source drone simulator AirSim. General Robotics co-founders Sai Vemprala (CTO) and Shuhang Chen came from the same Microsoft team.

GeekWire covered the launch in 2023, when it was Scaled Foundations and billed itself as “ChatGPT for robots.” It had five employees at the time, focused on aerial robotics and drones, with backing from Khosla and E14 Fund. It later renamed itself General Robotics.

Microsoft was sued Friday by the parent company of its hometown daily newspaper, The Seattle Times Co., which joined with Newsday to accuse the Redmond tech giant and OpenAI of using their journalism to train artificial intelligence models.

The lawsuit alleges that the companies scraped hundreds of thousands of Seattle Times and Newsday articles — bypassing paywalls and ignoring terms of service — to train their AI models. It seeks financial damages and the destruction of any training datasets and models built with their content.

“Like a snake eating its own tail, GenAI that is trained on painstakingly researched, expensive-to-produce content threatens to destroy the very news organizations by competing directly with them through AI-generated substitutive content,” the suit says. “If Defendants are allowed to succeed, independent journalism of the kind Plaintiffs produce will struggle to survive.”

The case is notable in part because the Seattle Times is suing two of its own funders. Microsoft Philanthropies underwrites some Seattle Times journalism projects. In 2024, Microsoft and OpenAI jointly funded a $10 million Lenfest Institute AI fellowship that included both the Seattle Times and Newsday among its inaugural participating newsrooms. The Times says it maintains editorial independence.

A Microsoft spokesperson said in a statement Friday evening, “While we’re surprised by the lawsuit, we appreciate the importance of the Seattle Times to our region and we’re always happy to sit down and explore solutions to this type of dispute.”

It’s not clear if there were negotiations or licensing talks in advance of the suit. GeekWire has contacted The Seattle Times Co. for comment.

In its own coverage of the lawsuit Friday evening, the newspaper quoted a memo from Seattle Times Co. President and CEO Alan Fisco, saying: “This was not an easy decision. However, we feel strongly that we must defend our content — which we spend millions of dollars a year to produce — from being used without our consent or compensation.”

The Seattle Times Union, which represents more than 160 newspaper employees, said Friday it supports the lawsuit but that in ongoing contract negotiations the company has refused to guarantee it won’t replace non-reporter newsroom jobs with AI.

“If the Seattle Times Co. truly cares about the threat AI poses to journalism’s business model, it should protect the workers who produce the copyrighted material at the heart of this case,” the union said in a statement.

Fisco, a longtime Seattle Times executive, took over as CEO on Jan. 1, succeeding Frank Blethen, who led the paper for 40 years and remains chair of the board. Ryan Blethen, Frank Blethen’s son and a fifth-generation member of the family that has owned the paper since 1896, became publisher in the same transition.

The complaint Friday includes examples of ChatGPT reproducing Seattle Times and Newsday journalism nearly word for word, including an 88-word verbatim stretch from The Seattle Times’ Pulitzer-winning coverage of the Boeing 737 MAX crashes, generated when a user prompted the chatbot with just the article’s headline and web address.

The suit echoes The New York Times’ 2023 copyright case against the same defendants, which just this week drew a U.S. Justice Department brief siding with Microsoft and OpenAI, arguing that a ruling for the publishers would stifle American AI development.

The newspapers join a growing list of publishers suing OpenAI and Microsoft over AI training. In addition to the New York Times, that includes the New York Daily News, Ziff Davis and the Center for Investigative Reporting, all consolidated before U.S. District Judge Sidney H. Stein in Manhattan.

On Friday, the publishers in that case moved for summary judgment, as did OpenAI and Microsoft.

OpenAI has struck licensing deals with more than a dozen other outlets, including The Associated Press, News Corp and Axel Springer. Publicly disclosed terms of three of those deals top $300 million, according to the Seattle Times complaint.

Updated with statement from The Seattle Times Union.

WSU President Betsy Cantwell is a full-on techie.

She spent more than two decades in national laboratories working on space exploration, energy and defense, followed by top leadership roles at state universities in Utah and Arizona driving significant growth in research funding. The longtime sci-fi fan embraces “future-casting” — a systems engineering approach to imagine and prepare for technological shifts a decade out.

But if that leads you to predict that the head of Washington State University is blanketing her campuses with artificial intelligence, you would be wrong.

The essential role of a university is to foster a sense of wonder in students, and that is something AI will never do, Cantwell said in a GeekWire interview on Friday.

That wonder, she said, “is innately human, and when we bring that into everything you do in a university, you create humans that will actually be much better at ideating, managing and creating value out of AI.”

In line with that philosophy, WSU faculty decide individually how to incorporate AI into their research and classrooms, with students actively involved in shaping its use. The technology serves as a tool to reach a goal, but is not the goal itself.

As the head of WSU — which serves 25,400 undergraduate and graduate students across its main campus in Pullman, four branch campuses, and an online program — Cantwell uses AI as a partner. On a $1.3 billion annual budget that also supports 39 county extension offices and research centers, she looks for practical AI applications to streamline and better inform her work.

Drawing on her national lab experience, Cantwell believes AI can bridge the gap between basic research and commercial production. Instead of universities simply “throwing papers over the transom,” she said AI-driven modeling helps build proofs-of-concept and testbeds that help attract private-sector investments.

This approach is at the foundation of WSU’s planned $50 million, federally funded research center using AI to support salmon recovery and sustainable aquaculture, as well as AI-enhanced micro-climate tools from the university to aid Washington’s farmers.

One area where Cantwell is personally using AI is in navigating the turbulent business of college athletics.

In 2023 — two years before Cantwell joined WSU — the University of Washington pulled out of the Pac-12, triggering an implosion of the century-old conference that left only WSU and Oregon State as members.

Cantwell believes strongly in the value of WSU’s athletics program, calling it a vital convener that brings alumni and fans together across political and social spectrums, while also providing an academic doorway for student-athletes who might not otherwise pursue higher education.

To navigate the financial fallout, she built her own custom AI agent to run tabletop scenarios and model financial paths for WSU’s sports enterprise. She treats the exercise much like national security strategic planning.

The AI agent allows her to analyze variables including coaching salaries, stadium operations, student-athlete employment litigation, federal regulatory proposals, and revenues generated through Pac-12 Enterprises, the conference’s business arm. Cantwell sees the grit of WSU rebuilding after the Pac-12 split as a compelling story of innovation and entrepreneurship, one that new digital media channels can help broadcast.

Where Cantwell draws the line, however, is AI use in sports betting, which she views as a concerning manipulation of human behavior that threatens the integrity of college athletics.

GeekWire came prepared for our interview by asking different AI chatbots to predict the outcome of Sunday’s Apple Cup showdown between WSU and UW at Seattle’s Husky Stadium. Suffice to say, the AI says it doesn’t look good for the squad from the Palouse.

But before we could deliver that news, the top Cougar made it clear she wasn’t interested in hearing what Gemini, Claude or any other AI platform forecast for the game.

“The only reason why I would want to know the score in advance is if I was betting and I don’t do that,” Cantwell said. “Why do you need to know in advance?”

Robotics jobs in the Pacific Northwest are multiplying faster than universities can train people to fill them. The University of Washington thinks it has an answer — or at least a start.

This fall, UW’s College of Engineering will launch its first robotics graduate programs: a Master of Science in Robotics and a Graduate Certificate in Applied Robotics. 

The university capped enrollment at 35 students for the inaugural cohort. More than 30 people had already signed up for an information session before applications even opened. It’s an  early signal, engineering leaders say, of pent-up demand from regional powerhouses racing to hire engineers who can operate at the intersection of AI, software and hardware.

Program leaders say that skill set is rooted in a traditional, narrowly focused engineering degree that hasn’t kept pace with the rapid evolution of technology. UW is betting that the fix lies at the intersection of AI and hardware, echoing an industry buzzword called “physical AI.”

“Robotics is no longer confined to a single discipline,” said Xu Chen, a UW engineering professor and director of the Boeing Advanced Research Collaboration, who played a large role in the committee that designed the new programs. “The future will need a wide variety of robotics knowledge, and that’s what we built these programs to deliver.”

Applications opened Sept. 1 and will close Sept. 10, with UW aiming for a roughly one-week turnaround before notifying applicants. For its inaugural year, the university is intentionally keeping things small: 25 seats in the master’s program and 10 in the certificate track, which is designed for working professionals who want robotics training without leaving their jobs.

Chen said the small first cohort is by design, not a limitation. The goal, he emphasized, is to get the fundamentals right before scaling up. He expects the programs to roughly triple in size within three to five years.

“Companies are seeing newer potential in robotics as advanced computing and the wave of AI technology mature,” Chen said. “They see that their workforce will benefit from a modern robotics program, and that need is really what drove this.”

Getting there will take machines, and lots of them. UW is purchasing robots and computing hardware for its initial course offerings while also leaning on industry donations: robots, GPUs, and computing infrastructure among them, according to Chen.

Amazon and Microsoft anchored the effort early; the list of partners has since grown to include NVIDIA, Boeing, Dassault Systèmes — the French software company behind design tools like SolidWorks — and at least one smaller robotics manufacturer.

“The industry board was incredibly supportive from the start,” Chen said. “We’ve had almost a year of continuous meetings and collaboration with them and with representatives across our own engineering departments.”

The broader structure of the program is meant to make it easier for students from different corners of engineering, such as electrical, mechanical and computer science, to land in the same classroom and eventually choose their own path deeper into robotics through electives.

The program is also drawing on UW’s existing research muscle in the region. It taps directly into the Boeing Advanced Research Collaboration, which Chen directs, along with robotics labs inside the Paul G. Allen School of Computer Science & Engineering. That gives students a line into the same research infrastructure that already feeds Seattle’s aerospace and e-commerce giants. 

The university’s ambitions extend well past this fall’s launch. Chen said UW has already mapped out longer-term plans for an undergraduate robotics degree and, eventually, a Ph.D. program, with the two new offerings serving as the foundation.

For now, the clearest sign of the program’s ambitions arrived this summer in an unlikely form: a pack of robot dogs let loose on UW’s campus.

“Seattle’s hills make it a uniquely difficult place for robots to move around, which is exactly why it’s a great place to study it,” Chen said.

Both students and faculty got a chance to operate the robots directly, Chen said. It was a hands-on moment that underscored how much more accessible robotics technology has become in just the last few years.

“It was exciting to see the students so happy to see the robots,” Chen said. “That’s the kind of energy we want to build this program around.”

We’ve come a long way from Lincoln Logs and Hot Wheels that couldn’t talk to us. Today, plush toys aren’t just stuffed — they’re stuffed with artificial intelligence, and new research from the University of Washington reveals that when these “smart” toys start chatting, kids quickly go from curious to frustrated to outright hostile.

Claims of “smart” toys date back decades, from 1960s talking dolls like Chatty Cathy to 1990s sensor-packed plushies like Microsoft’s ActiMates Barney and Furby.

But generative AI marks a major shift. Companies like Curio are now packing plushies with onboard AI models, allowing characters like “Gabbo” or the viral brainrot figure “Ballerina Cappuccina” to hold dynamic, unscripted conversations, remember past interactions, and adapt directly to a child.

To see how kids actually interact with these conversational companions, researchers at UW’s KidsTeam brought eight children ages 6 to 11 to campus last summer. The kids initially engaged with curiosity — asking basic questions like “What is your name?” and testing physical reactions like tickling the toys’ toes.

But as the toys struggled with complex questions and failed to pick up on physical cues — one participant complained a toy “didn’t listen to me like 26 million times” — delight turned to irritation. Children eventually turned to antagonizing the plushies, calling them “ugly” or “evil” and joking about throwing them in the ocean.

In the video below, kids are asked at one point if they want an AI toy to read them a bedtime story.

“No. It just sounds awful,” one child replied.

“I think it’s gonna destroy my dreams as a tiny kid,” another said.

The study highlights a distinct psychological clash: a cuddly, familiar plush exterior combined with a synthetic intelligence that kids found both fascinating and unnerving.

“The juxtaposition of this plushie toy that also had signs of intelligence was both interesting and disturbing for the kids,” said co-lead author Aayushi Dangol, a former UW doctoral student now at Foundry10, in a UW News story.

While the toys offer dynamic play, Dangol warned parents that generative AI introduces new risks that traditional toys never had, from hallucinating facts to manipulative emotional bonding.

“They’ll give wrong answers, or flatter the kids excessively, or could manipulate the kids into attachment,” Dangol noted.

Beyond conversational glitches, researchers emphasize that synthetic companions fundamentally alter how children play. For generations, kids have supplied their own imagination to make inanimate objects talk and move. Generative AI alters that dynamic.

“Now the script has been flipped and the toy has this imitation of imagination,” said co-author Jason Yip, a UW associate professor in the Information School and director of KidsTeam UW. “We’ve never lived through that before, and we don’t know what questions children will ask or how long they’ll even want to play with these toys.”

Because kids are navigating entirely uncharted territory, Yip stressed the importance of giving young users space to talk through their experiences with the devices filling their bedrooms.

“It’s really important to give them opportunities to discuss these technologies we’re handing down to them,” Yip said.

Seattle-based venture capital firm Madrona released its sixth annual Intelligent Applications 40 list this week, naming 45 private AI companies (the five extras come from ties) that have collectively raised $410 billion from investors across the industry.

Three of them — Anthropic, OpenAI and Databricks — account for 92% of that total.

The uneven distribution of funding reflects a larger split in the tech industry, as the largest AI companies make huge bets on the computing capacity needed to meet demand for their models, while almost everyone else builds businesses on top of them.

The frontier labs are “increasingly funded by strategic capital from the likes of Amazon, Google, Nvidia and SoftBank rather than traditional venture,” Madrona’s Matt McIlwain and Rolanda Fu wrote in a post accompanying the list. That scale, they added, “makes every other category on this list look capital light by comparison.”

On top of that, he said, hundreds of billions of dollars are flowing into OpenAI and Anthropic.

“And what I say to both the big tech companies and to the people funding the model companies: thank you very much,” McIlwain said on Bloomberg TV, noting that the five largest tech companies will spend an estimated $750 billion in capital expenditures this year.

But even setting those big three aside, McIlwain said, the rest of the winners have raised an average of more than $800 million each. That’s a total of $34 billion combined. Companies across the list are raising far more than they used to, enough that Madrona had to redraw its own categories.

The list sorts companies by total capital raised, and this year the ceiling for “early stage” rose to $50 million, up from the $30 million threshold that held for the previous five lists. The cutoff for “emerging enablers,” its category for smaller infrastructure companies, doubled to $100 million.

“Companies across the board are raising more money, and the definition for what ‘early’ means continues to shift higher,” McIlwain and Fu wrote.

Madrona has published the IA40 since 2021 as a roster of the private companies it considers most important in building and enabling AI applications. According to the firm, this year’s list drew on input from 72 investors representing 54 venture and corporate firms, who nominated and voted on more than 450 companies, with PitchBook data factored into the scoring.

Two Seattle-area companies made this year’s list:

• Clarify, which builds an AI-native CRM, appears in the early-stage category for the second straight year, with $23 million raised. It acquired San Francisco’s Seam AI in July.
• Gradial, which builds AI agents for enterprise marketing, is a first-time winner in the mid-stage category; it raised $65 million in June at a $675 million valuation, bringing its total to $120 million.

Last year’s list included two other Seattle-area companies in addition to Clarify.

• OpenAI acquired one of them, Bellevue-based Statsig, for $1.1 billion in September 2025, making Statsig founder Vijaye Raji its CTO of applications.
• Security startup Dropzone AI, which was on the list last year, did not repeat this year.

Madrona, one of the Seattle region’s largest and oldest venture capital firms, is an investor in all four — Clarify, Gradial, Statsig and Dropzone AI — although it also invests outside the region, and many of the companies on the IA40 are not in its portfolio.

Several of the companies on this year’s list have engineering centers in the Seattle region, including Anthropic, which leased 113,000 square feet in South Lake Union this year; OpenAI, which expanded to nearly 300,000 square feet in downtown Bellevue after the Statsig acquisition; and Anduril, which employs about 560 people in Bellevue and Seattle.

Databricks, the San Francisco-based data and AI company (which leased 142,000 square feet in Bellevue this year), is the only company to appear on all six IA40 lists. That said, 23 of last year’s 40 winners returned this year, a 58% repeat rate, up from 33% the year before.

McIlwain and Fu wrote that the biggest and most established companies on the list are holding their spots, noting that “the age of experimentation is giving way to an age of enterprise readiness,” with buyers and investors “paying premiums for companies that can demonstrate real ROI.”

Madrona will recognize the winners at its IA40 Summit in Seattle on Sept. 29 and 30.

Updated with Matt McIlwain’s comments to Bloomberg TV.

The co-founders of Pro.com, the Seattle-based home-improvement marketplace acquired by Opendoor in 2021, are back with a new company targeting what seems on the surface a very different kind of market: AI-powered software for the wealth management industry.

But Rajalakshmi “Raji” Subramanian and Matt Williams say the new challenge matches the same pattern: a huge industry held back not by a lack of customers, but by a shortage of professionals and tools.

Their Seattle startup, OnTrade, co-founded with former Bank of America chief investment officer Zachary Harl, has been operating under the radar since 2024, raising an undisclosed amount of funding from General Catalyst, Madrona and angel investors.

OnTrade’s chief technology officer is Jean Bredeche, who co-founded Quantopian, the algorithmic trading platform, and later served as a director of engineering at Robinhood.

How it works: OnTrade connects software that financial advisors already use — including CRM, portfolio accounting, trading, and compliance programs — into a single interface.

It then deploys AI agents to handle the type of work that advisors have traditionally done manually, such as scanning portfolios for tax-loss harvesting opportunities, flagging accounts that have drifted from their targets, or drafting proposals and reports for clients.

The humans approve everything before it reaches a client. The idea is to help them serve more clients without sacrificing the quality of their work, expanding access to wealth-management services that tend to be concentrated among more affluent households.

“Wealth management, if you look at the industry, does not have a demand problem; it has an access problem,” said Subramanian, the company’s CEO, in an interview. “Many people who’d like access to wealth management don’t have access to wealth management, and that’s what we’re here to solve.”

Harl, OnTrade’s chief investment officer, called raw foundation models the “brilliant PhDs” of the AI world — impressive on paper, but not as valuable to a specific industry such as wealth management until they understand its portfolios, policies, compliance rules, and client relationships. Vertical AI solutions like OnTrade, he said, are better positioned to connect that general-purpose intelligence to a specific firm’s data and workflows so the technology can do trusted work.

Industry shakeup: OnTrade is emerging at a pivotal moment, two days after investment giant Vanguard agreed to acquire wealth-management platform Altruist reportedly valued at $4 billion. OnTrade’s founders cite the deal as validation of the vertical AI opportunity they’re pursuing.

In a LinkedIn post Thursday, Subramanian wrote that the Vanguard-Altruist deal signals something bigger than a battle over where advisors park their clients’ assets: that capturing the opportunity “requires a new operating model rather than AI-enhanced versions of today’s applications.”

The wealth management industry’s unit of scale, she wrote, is shifting “from the number of people a firm employs to the intelligence and agency it can deploy.”

The founders: Subramanian joined Amazon in the late 1990s as an early engineer who helped build Amazon Marketplace and AWS, and later led the digitization of books for Kindle.

Amazon was where she met Williams, who had founded a startup called LiveBid that Amazon acquired in 1999. He spent 11 years there, including a stint as a technical advisor to Jeff Bezos, then left to run Digg as CEO and served as an entrepreneur in residence at Andreessen Horowitz.

Subramanian went on to run engineering at Yahoo Finance, where she helped open up market data that had previously been the province of institutional investors, giving her an early look at the problem that OnTrade is now aiming to solve.

In 2013, the two co-founded Pro.com, a tech-driven home improvement marketplace that raised early funding from investors including Madrona, Maveron, Bezos and Andreessen Horowitz.

Real estate tech company Opendoor acquired Pro.com in 2021, and brought both founders on as executives — Subramanian as chief technology officer, Williams as head of the Pro.com unit and senior vice president of retail.

Harl spent many years at Bank of America, rising to chief investment officer, where he managed the bank’s asset portfolios and large balance sheet risks across multiple market cycles. He is a chartered financial analyst (CFA), with a math and computer science degree from Indiana University, and a statistics degree from the London School of Economics.

He served on the U.S. Treasury Borrowing Advisory Committee under Secretaries Steven Mnuchin and Janet Yellen, advising on debt management, before joining Opendoor in 2023 as chief risk officer. That’s where he met Subramanian and Williams, before making the startup leap with them.

Traction and competition: The company’s technology is already in use at firms ranging in size from boutique advisories to large national practices, said Williams, the company’s president.

He said one client used the platform to win a billion-dollar family office account, and that another recouped the full annual cost of the platform in less than 30 days. He called that “a small window into what’s going to happen on a larger scale.”

The wealth management software market has many established players — such as Orion Advisor Solutions, Envestnet, and Addepar — but the OnTrade founders say they see them as partners, not rivals. OnTrade integrates with those systems rather than replacing them.

That distinguishes the company from Altruist, the Vanguard acquisition target, which built its own full stack, including its own custodian, the financial institution where client assets are held. That approach requires firms to move client assets onto its platform.

OnTrade doesn’t ask firms to replace their existing tools or move their clients’ money. Instead, it plugs into what’s already there.

The broader timing may work in their favor. As baby boomers age, an estimated $50 trillion or more in assets is expected to pass to younger generations in the coming decades — creating a wave of new clients who will need financial advisors, and new pressure on firms to serve them.

That’s where home improvement and wealth management have something in common.

“There aren’t many bigger places, other than health, wealth and real estate, where you can impact a population, especially an underserved population,” Williams said. “That was at the heart of the motivation.”

An AI system built by Seattle’s Allen Institute for AI (Ai2) has found evidence that a common form of breast cancer, long thought to be a poor candidate for immunotherapy, might actually respond to it.

The finding, produced by Ai2’s AutoDiscovery system, has led to an expanded partnership with the Paul G. Allen Research Center at Providence Swedish Cancer Institute, which is now deploying the AI system on its own patient data to look for similar scenarios across other types of disease.

The announcement Thursday illustrates the broader potential for AI to uncover findings that human researchers, overwhelmed by the massive scale of modern datasets, might otherwise miss.

“Cancer researchers have access to extraordinary datasets, but the challenge is no longer collecting data; it’s understanding everything those datasets have to tell us,” said Dr. Kelly Paulson, who leads the Center for Immuno-Oncology at the Paul G. Allen Research Center, in a statement.

AutoDiscovery, announced by Ai2 in February, works differently from most AI research tools: instead of waiting for a scientist to pose a question, it starts with a dataset and generates its own hypotheses, ranking them by how much they challenge existing assumptions.

In a research paper posted to the preprint server MedRxiv, the Ai2 and Providence Swedish researchers explain that they applied AutoDiscovery to The Cancer Genome Atlas, a federal dataset spanning more than 30 types of cancer.

The system flagged signs that invasive lobular carcinoma, which accounts for about 15% of U.S. breast cancer diagnoses, may be more responsive to immunotherapy than researchers thought. It has largely been left out of immunotherapy trials.

The Providence Swedish team confirmed the finding in a second patient dataset and validated it in tumor tissue in the lab. They cautioned that the findings don’t prove immunotherapy would work in these patients but suggest the question warrants further study.

The collaboration connects two organizations that trace their origins to the late Microsoft co-founder Paul Allen, who founded Ai2 in 2014 and whose $20 million donation helped establish the research center at Swedish in 2024.

GeekWire is profiling over the next few weeks some of the people and teams that are shaping the evolution of Microsoft in what we’re calling its “Microsoft 2.5” era.

From AI Frontier Lab to Frontier Ecosystem: Microsoft got a foothold in AI thanks largely to its partnership with OpenAI. But that’s not the way it is planning to continue growing its AI business.

Inside Microsoft AI (MAI), the Microsoft Superintelligence team is focused almost entirely on building its own frontier-level models. That team already has developed a handful of home-grown offerings, including MAI-Code-Flash for writing code faster; MAI-Cyber-Flash, a cybersecurity model; and MAI-Image, a model for creating images.

The head of the Superintelligence team is Ali Farhadi, corporate vice president of AI. Farhadi, who joined Microsoft five months ago, is also a professor at the University of Washington, where he has worked for nearly 15 years. He was previously CEO of the Allen Institute for AI (Ai2) and before that was an AI and machine learning leader at Apple for more than three years, after it acquired his startup, Xnor.ai.

When he joined Microsoft, Farhadi said in a LinkedIn post that he believed “Microsoft has all the pieces to win in this AI race: data, search, coding, infrastructure, agents, software and the world’s biggest Fortune 500 companies taking dependencies on Microsoft every day.”

Farhadi elaborated on that in an interview with GeekWire this week. AI is shifting from a “Frontier Lab” era to a “Frontier Ecosystem” era, he said. It’s no longer just about training models; it’s about integrating the models with enterprise data, platforms, distribution systems and customers in a trusted way.

The next battlegrounds in AI will be around cost, reliability, specialization, and deployment at scale, rather than simply building larger models that beat others in benchmark scores, he said.

“If you look around, there are not that many places to have all these missing pieces together at scale, especially if you add the element of trust to it,” Farhadi said.

Cutting through the AI noise: Farhadi said his management philosophy is grounded in the importance of personal relationships, which are especially key in big organizations. People need to understand your rationale and to trust you can deliver on what you’re tasked to do, he said — an approach that has served him inside both Microsoft and Apple.

Staying on top of the flow of information while filtering out the AI noise makes prioritizing crucial. The team has “a long list of things that we believe we should be doing,” he said, but much of it stays on the back burner to maintain a “laser focus on delivering on the main mission.”

The priority is building high-quality models, both generalist and domain-specific. On the domain-specific front, Microsoft is working with the Mayo Clinic on a healthcare-specific model based on Mayo’s own clinical data, as well as Microsoft’s cybersecurity and coding models.

The thinking: For a lot of enterprise work, a narrower model beats a bigger one.

“If you can do something at [the same] quality or better quality at a fraction of a cost, it’s just a no-brainer. And having a way to specialize to domains, to industries, to enterprises is one way,” he said.

Microsoft execs have referred to this approach as a “hill-climbing machine,” meaning the ability of a model to scale and continuously improve within a specific domain. Microsoft is coupling the hill-climbing with “frontier tuning,” like it is doing with the Mayo Clinic. Frontier tuning includes customizing frontier models; keeping proprietary data private, preserving institutional know-how; and avoiding leaking intellectual property (IP) into shared models.

“We all thought that IP is your data,” Farhadi said. “But we learned that IP is also how you work.” And that’s why safeguarding these elements is so crucial.

Open all the things? Farhadi led an expansion of open-source AI development at Ai2, the Seattle-based institute founded in 2014 by the late Microsoft co-founder Paul Allen. While Microsoft has contributed to the open-source community on various fronts, including AI tooling, it hasn’t open-sourced its frontier models.

Farhadi said he personally remains “a big advocate of open source,” but noted that the industry has changed since his Ai2 days as there are now more credible Western open-source models and businesses forming around them.

He didn’t rule out Microsoft doing something in open-source models, or the somewhat less-open “open weights” area, but there’s seemingly nothing happening on that front in the near term.

In the coming months and beyond, the focus of Farhadi’s team is helping Microsoft turn into a Frontier Ecosystem by building cutting-edge AI capabilities; helping enterprises create their own tuned versions of them; continuously improving models; and making sure customers keep control of their own destinies and data.

Success for Microsoft’s Superintelligence team has nothing to do with the idea of Artificial General Intelligence (AGI) which OpenAI, Anthropic and others have positioned as their ultimate goal over the years. In fact, when I asked Farhadi about AGI, he said, “I don’t understand what that means.”

Don’t worry, Ali. You’re not the only one.

When Bill Gates talks, people listen. This week he published a lengthy essay on what AI is going to do to work, and told GeekWire that people inside AI companies who name the downsides get told, “Hey, you’re hurting our PR while we’re trying to raise trillions of dollars.”

He’s right about the hard part. The job displacement he describes lands on young workers first, and the safety net is funded by taxes on the very wages that AI erodes. He prescribes three treatments: new institutions at home and abroad, a tax on AI tokens and robots, and “Human Reserved,” a category of jobs only people may hold.

Gates has the diagnosis right but the prescription mostly wrong. I’d sign the robot tax tomorrow, because hiring a person costs you payroll tax every year while buying a robot gets written off in year one. The other two I’d send back.

Let’s start with what’s solid. Stanford’s Digital Economy Lab updated its “Canaries in the Coal Mine” work this month. Employment for 22-to-25-year-olds in the most AI-exposed occupations is running 19% below where it would be if it had kept pace with their peers in less exposed work, up from 15% a year ago. The same authors say they don’t see widespread, economy-wide displacement, and unemployment held at 4.1% in July.

The AI damage isn’t arriving as layoffs. It’s arriving as jobs that never get posted, and Gates is right that the young get it first.

Now the token tax. Tokens (essentially words) are what AI companies bill by. Taxing tokens is like taxing keystrokes: it measures effort, not displacement.

A high school class working through calculus with an AI tutor burns tokens continuously. A model that quietly retires a 40-person customer center might burn relatively few. The tax lands hardest on the uses Gates says he wants to protect.

Stanford’s AI Index put the cost of GPT-3.5-level performance at $20 per million tokens in November 2022 and seven cents by October 2024, a 280-fold drop. You’d be indexing the safety net to a number that falls every year while displacement rises.

And you can’t collect it. Inference runs on laptops and phones now, and on servers in whatever country declines to sign. A token tax is a tax on whoever uses an American API, and every dollar it adds makes a Chinese model look cheaper. We’d be slowing ourselves down and not China.

Gates says the institutions will take years to build, and also says we can’t afford to move slowly. He’s right twice, and that’s the problem. He wants the international body to borrow from nuclear inspections and aviation regulation. That may pan out in the long term, though the UN is the cautionary tale for the bureaucratic nightmare that the international community can produce.

Meanwhile we have functional agencies with jurisdiction today. The FDA can rule on AI in diagnosis. The FTC can go after AI-enabled fraud. We don’t need a new agency to say a bank can’t deny your mortgage because a model felt like it. We need the banking regulator to reiterate it forcefully.

That leaves Human Reserved, his best idea but his most privileged one. Gates would protect a job for either of two reasons: the role is deeply personal, like a caregiver, or the people who hold it are unlikely to find other work. Only one of those holds.

Freezing headcount because the workers have nowhere else to go protects the job for a while and makes the service more expensive along the way. Reserving the moments when a human being is the point is defensible, and Gates makes that case well. On a robot delivering the news that you have an incurable disease, he writes, “There’s no technical reason why it couldn’t,” and adds, “Yet it shouldn’t.” He’s right.

I made the case in WIRED nine years ago that displaced workers should move into caregiving, and that it would take real money to lift the pay enough to draw them.

The problem with Human Reserved is that it assumes there’s a human being available. Home health and personal care aides earn a median of $34,900 a year, and BLS projects roughly 765,000 openings in that occupation every year through 2034. At that wage, they keep coming open. A third of home care aides are immigrants, and tighter enforcement threatens that supply. A rule that reserves care for people, in a market with no spare people, reserves care for the families who can outbid everyone else.

Gates half-anticipates this, telling The New York Times he might be a flawed messenger because of his wealth. On this point he is. The caregivers who gave his father something irreplaceable were in that room because someone could pay them to be there.

So don’t fence AI out of the room. Put it to work in the hours nobody is paid to cover.

In February the Times ran Eli Saslow’s story about Jan Worrell, 85, living alone on Washington’s Long Beach Peninsula with an AI companion called ElliQ that engages her about eight times a day and pushes her to stay hydrated and moving. (I serve on ElliQ’s board, and I joined because the company builds a machine that extends a caregiver’s reach instead of replacing one.)

Her goal, she told her doctor, was to never live anywhere else. Fund enough aides to cover the hours that need a person and put the machine on the rest.

Here’s where I net out: equalize the tax treatment of labor and capital, which Congress could do next session, and route the proceeds into retraining and into topping up the pay of workers who land in lower-paying jobs. That’s a better answer than a protected job title.

Drop the token tax, build the caregiving workforce instead of fencing it off, and use the regulators we already have while somebody works on the ones we don’t.

Bill Gates is legendary, bordering on notorious, for his late-night emails — missives to colleagues with piercing questions about Java back in the day, or malaria these days, or whatever esoteric topic he happens to seize upon at any given moment.

But increasingly, he is sending these messages to AI, not to people. He’ll bounce something off Claude, get ChatGPT to weigh in, and insert himself in the middle.

He described the pattern in an interview with GeekWire: “It’s 3 a.m., I want to understand sodium batteries. Now, there’s no reason to go to sleep. Here we go! Yeah, it’s crazy.” 

If you’re a curious person, he said, “this is a mind-blowing time.”

In terms of productivity, he added, “we are in heaven.”

All of which might be predictable. This is Bill Gates, after all. Now 70 years old, he has spent more than five decades impatient for the future to arrive — making the case that innovation, on the whole, will ultimately put humanity and the world in a better place.

So here’s the surprise twist: He’s now deeply concerned about where technology is headed, how fast it’s progressing, and how little the world is doing to get ready.

In a new essay, Gates says the “turbulent AI era” has arrived, with technology threatening to erase categories of jobs, supercharge fraud and deepfakes, lower the bar for cyberattacks on critical infrastructure, make it easier to engineer a deadly new disease, let governments kill without humans involved in the decision, and fundamentally change how kids grow up.

If someone came up with a credible plan to slow the pace of AI globally, he writes, he’d likely support it. But he doesn’t expect one. The geopolitical and economic forces are too much. 

He says that the world needs to take action, and offers three ideas to start:

Build new institutions, at home and globally. No existing agency was designed for a technology that touches jobs, security, health, energy and elections all at once, he writes.

Gates calls for new national bodies that can set priorities across agencies, plus a new international organization modeled on nuclear weapons inspections, aviation rules and the ozone treaties.

Set aside jobs for humans. Gates calls this “Human Reserved”: work that machines will be fully capable of doing, but that we decide to keep for people anyway. The model is a nature reserve — land where we could build roads and buildings, but choose not to, because the loss would be too great.

One example: a robot delivering the news that you have an incurable disease. “There’s no technical reason why it couldn’t,” he writes. “Yet it shouldn’t.” 

The idea came in part from watching the caregivers who looked after his father through Alzheimer’s, work he describes as “irreplaceably human.” 

Tax AI tokens and robots. Today a company that hires a worker pays payroll taxes, while a company that buys a robot deducts the cost. Gates says that gives employers a reason to replace people. He’s calling for a tax on AI to change the incentives and help pay for retraining. 

He first floated a robot tax nine years ago, but the idea was widely dismissed. He’s still for it. He acknowledges that it isn’t economically efficient, but says that with innovation accelerating, we can afford a little inefficiency as the price of keeping people employed.

Gates is candid that he doesn’t have all the answers, particularly on the proposal for “Human Reserved” jobs. Who decides what gets reserved, and by what criteria? How do you keep companies from using robots in the jobs that are supposed to stay human? 

These, he writes, “will need to be worked out in public.” 

In the meantime, he’s working it out with Claude. Gates said he has talked the idea through with the chatbot, thinking through different ways to get the share of work reserved for humans up to 40%, using shorter workdays and earlier retirement to spread what’s left around.

Crossing the threshold

In the GeekWire interview, Gates said the essay came out of a specific realization: the AI industry is blowing past its own warning signs, one after another, and almost nobody is saying so out loud. 

For years, he said, people in AI described certain moments as dangerous points where the industry would stop and think hard before going further: making it easier to build a bioweapon, making it easier to launch a cyberattack, building machines people become emotionally dependent on, wiping out large numbers of jobs, and losing control of the technology itself.

“We’re in the process of crossing every single one of those thresholds,” he said.

Meanwhile, nobody in the industry wants to be first to step on the brakes. “Most people you talk to will say, yeah, well, if everybody else would slow down, maybe I would, too,” he said.

Gates said one way out of that standoff is for governments to step in. 

His example: any AI model capable of designing new molecules — the capability that would let someone engineer a new disease — should be monitored. The monitoring would be mandatory rather than voluntary, and it would cover free models as well as commercial ones. It would also have to be written so a company can’t copy the model elsewhere and strip the monitoring out.

“To me, that’s kind of like common sense,” he said. “But we don’t see a specific proposal to do that.”

‘The whole thing seems so empty to me’

Under an executive order signed by President Trump in June, AI companies are asked to submit their most powerful models for government testing up to 30 days before release. The order specifically bars the program from becoming a licensing or preclearance requirement. The White House finalized the framework in early August.

Gates said he doesn’t get it.

“What is the threshold that’s being examined, and what is the action taken when you cross that threshold?” he said. “The whole thing seems so empty to me.”

If the world can’t take these basic steps, he said, “I really am going to throw up my hands.”

If the process stays voluntary, with no line and no consequence for crossing it, “we’re going to look back on this as a kind of eye-of-the-storm type moment,” he said.

Asked if he had taken his proposals to the Trump administration or to other heads of state, Gates said with a bemused tone, “Well, you could tell me who at the White House I should be talking to about this.” He said he hopes the essay reaches people in Congress and in the executive branch.

He said the public argument among AI companies over whether the risks are real is beside the point, because privately the people running them already agree. “I know they’re all worried,” he said, “or all of them that I know, which is basically everybody but Elon.”

People inside AI companies who acknowledge the downsides, Gates said, get told: “Hey, you’re hurting our PR while we’re trying to raise trillions of dollars.”

Gates said he previously expected losing control of AI to be a distant problem, something to worry about “many years from now.” He’s no longer convinced that’s the case.

He referenced an Aug. 11 episode of the Dwarkesh Patel podcast featuring Ryan Greenblatt, chief scientist at the AI safety group Redwood Research. Greenblatt said that as AI systems get more capable, the people building them understand less and less about what is happening inside, and that sufficiently advanced models could end up working against their creators.

“These are people who are super expert on the thing, going, well, maybe we won’t be able to control these things,” Gates said. “I mean, what kind of risk have we chosen to run here?”

In the poorest countries, he expects AI to do more good than harm. In the countries where the Gates Foundation works, doctors, teachers and farm advisors are all in short supply. AI can help fill those gaps. The foundation will lay out that work at its Goalkeepers event next month, including an effort to make AI models work as well in African languages as they do in English.

The job losses, he added, will hit rich countries first.

Gates published the essay early Wednesday morning, and it’s drawing coverage from a variety of outlets, including The Wall Street Journal, the New York Times, and MIT Technology Review.

It’s the first big wave of new attention on the Microsoft co-founder and Gates Foundation chair since he answered lawmakers’ questions in the Jeffrey Epstein investigation on June 10, sitting for a nearly six-hour voluntary interview with the House Oversight Committee.

Gates, who has not been accused of any wrongdoing, was asked by Axios whether he’s concerned that the Epstein issue could undercut his message. According to the site, he compared this to earlier situations when personal and professional challenges diminished his ability to speak out on key subjects: during the Microsoft antitrust trial, and his divorce from Melinda French Gates.

The AI Road Ahead

For all of this, Gates is still thinking about how technology will change human life and productivity, in many ways for the better on an individual level. 

• A key step, he said, will be establishing broad-based persistent memory for AI agents across contexts. For now, AI still doesn’t know you like a human assistant who’s familiar with your relationships and how you think about your time. 
• Gates sees the role of apps changing in the future. Instead of bouncing between different pieces of software, he said, AI will increasingly be the primary interface. “You won’t go to those applications,” he said. “You’ll just go to your personal agent.”
• He also sees AI continuing to transform shopping, to an extreme: “We will get to a point where you won’t buy things yourself. You just won’t.” Telling the agent to help you buy something, “it’ll consider so many more things, and it’ll make it so much easier for you to do it.”

Asked whether he is still an optimist, Gates didn’t answer directly. “I don’t think being pessimistic is helpful,” he said.

“I do think, wow, this is sure an interesting time. I’m the guy who in my 30s thought people in their 50s or 60s didn’t understand anything.” He called it “kind of bizarre” that he would be delivering a message like this at 70.

“But I am very concerned. And honestly, when you get people one-on-one, so are they.”

The next big test for AI isn’t happening in a data center. It’s happening in the dirt.

Really, it’s in the weeds.

Two Seattle-area startups are betting that AI can transform how farmers find and eliminate unwanted plants — one by mapping every weed, the other by zapping them with lasers.

Issaquah, Wash.-based TerraClear is commercializing a new system that uses ultra-high-resolution imagery and machine learning to map individual weeds across entire fields of corn and soybeans, then turns those detections into digital prescriptions that can be sent directly to precision sprayers.

The new Weed Maps technology from TerraClear — best known for its robotic rock picking technology — can identify weeds as small as a quarter of an inch, the company said in a press release today.

Meanwhile, Seattle-based Carbon Robotics is taking a different approach: Its autonomous LaserWeeder uses computer vision to identify weeds and then blasts them with lasers.

Now, the AI powering these systems is getting smarter, too — moving beyond simple weed detection toward models that can recognize and understand plants across different crops, fields and growing conditions.

TerraClear’s new Weed Maps, announced Tuesday, captures imagery at 1.5-millimeter resolution and identifies weeds as small as the eraser on a pencil. Rather than sampling portions of a field, the company says it collects images of every acre and produces a geo-referenced map that can be uploaded to section or nozzle-controlled sprayers used by farmers.

The goal is precision at a level that would be difficult for a human to achieve, allowing a farmer to know where the individual weeds are.

TerraClear says the maps can be delivered the next day, giving growers a chance to act while weeds are still small and easier to control.

Devin Lammers, the chief executive of TerraClear, tells GeekWire that its approach “sidesteps the capital problem entirely.” In other words, farmers need not buy a new piece of expensive equipment, instead using software to turn existing sprayers into precision instruments by telling them exactly where to spray.

He called Carbon Robotics laser-weeding system “impressive technology,” noting that it works well for specialty crops and organics.

But bigger farms producing commodity crops like corn and soybeans — the market TerraClear is going after — need a different approach, he said.

“Modern grain and oilseed sprayers already have individual nozzle control and RTK positioning — the actuation hardware is sitting in the shed,” Lammers said via email. “We just hand the sprayer a shapefile of individual weed locations and it turns the nozzle on only where a weed actually is.”

Given that large corn and soybean growers farm more acres at a lower revenue per acre, Lammers said it’s a “very different P&L” where expensive new equipment needs to pencil out.

With TerraClear’s new system, Lammers added that “the farmer buys a map, not a machine.”

RFK Jr. and new ways to farm

One of the benefits of both approaches is chemical use reduction in the field, a hot topic in political circles with President Trump earlier this year committing $1 billion to modernize farming and reduce chemicals in agriculture. That federal investment could help spark new innovations, like the ones TerraClear and Carbon Robotics are developing.

Robert F. Kennedy Jr., the U.S. secretary of health and human services, earlier this year touted Carbon Robotics’s machines on an episode of The Joe Rogan Experience as a possible solution in cutting pesticide use.

In the case of TerraClear, Lammers said the precision mapping technology alone could cut pesticide and herbicide use by up to 80% with no loss of efficacy.

Both startups are part of a broader Pacific Northwest ag-tech ecosystem that has been applying AI and robotics to agriculture, building on the regions farming and tech roots.

TerraClear founder Brent Frei represents that unique farming and tech DNA. He grew up on a family farm in Grangeville, Idaho, before studying at Dartmouth and then moving to the Seattle area where he co-founded Onyx Software and Smartsheet.

Founded in 2017, TerraClear originally attacked a much less glamorous agricultural problem: identifying and removing rocks from farmers’ fields. In 2024, the company raised $15 million, bringing its total funding to $53 million.

By February of this year, TerraClear had expanded to about 50 employees and was approaching 1,000 customers. At that time, it also launched an autonomous field robot called TerraScout, designed to collect high-resolution imagery across a field and convert that information into actionable maps for existing farm equipment.

The company says TerraScout can collect more than 4 billion image samples per acre and map more than 1,000 acres a day under favorable conditions.

In addition to TerraScout, Lammers said they are using aerial drones to ingest field-level data into its new Weed Maps product.

“That’s the part that compounds — the imagery we gather is field-level, repeated season over season, and specific to the commodity acre,” Lammers said. “Models get better, which makes the maps better, which brings more acres, which produces more data.”

Carbon Robotics is further down the road in making the machine the decision-maker, and eradicating weeds without the use of chemicals.

The Seattle startup’s LaserWeeder combines cameras, AI and high-powered lasers to identify weeds and destroy them without applying herbicides or pesticides. The company has deployed its machines on farms around the world and has built an enormous dataset in the process.

Announced in February, its so-called Large Plant Model was trained on 150 million labeled plants, which Carbon describes as the largest agricultural plant dataset of its kind. The company’s goal is to move beyond narrowly trained computer-vision systems that need to be retrained whenever a new weed or field condition appears.

With the Large Plant Model, farmers can use Carbon’s Plant Profiles feature to show the system a handful of images and customize what the machine should recognize and target.

Given the changing dynamics of a weed during various stages of its growth — and based on conditions such as soil, weather and crop varieties — Carbon wants to correctly identify the difference between a weed and a crop.

“When our robots can understand any plant in any field immediately and adapt behavior in real-time, farmers immediately get maximum value from the machines,” Carbon Robotics CEO Paul Mikesell said in a press release. “The Large Plant Model provides farmers with the most advanced AI technology to maximize the weeding quality of LaserWeeder in their unique environments.”

Founded in 2018, Carbon Robotics has raised $177 million to date and as of last year employed about 260 people at offices in Seattle and a manufacturing facility in Richland, Wash.

The farm becomes a giant AI dataset

TerraClear and Carbon Robotics are attacking one of agriculture’s thorniest problems — weed management — from different directions.

TerraClear wants to allow a farmer to keep using a conventional precision sprayer, while making it dramatically more selective via its Weed Maps.

Carbon, meanwhile, is developing autonomous laser-weeding equipment itself, identifying and eliminating the individual weeds in real time without chemical spray or tractor operators.

The bigger opportunity for both companies may ultimately be neither maps nor lasers, but the underlying data they gather.

Every time a camera passes over a field, it can collect information about plants, soil, crop health and growing conditions. That’s vital information to farmers, seed producers, agriculture researchers and equipment manufacturers.

Visiting an Amazon delivery station can feel like walking into the past.

While many of its massive fulfillment centers are equipped with the latest robots and automation, Amazon’s delivery stations — the final stop before packages reach the doorstep — remain mostly manual. Workers often sort parcels by hand, load them into bags, and stage them for drivers.

That could be changing. Business Insider reports that Amazon is developing an internal initiative called Project Tetromino to build “fully automated” delivery stations, citing an internal planning document that includes specific financial projections.

The name appears to be a nod to Tetris, reflecting the puzzle-like challenge of efficiently organizing packages for delivery vehicles.

“We’re always exploring and testing new technologies across our operations to improve safety and the delivery experience for customers,” Amazon spokesperson Brad Glasser said in a statement. He added, “The details cited here are inaccurate and don’t reflect our current plans. Like any early-stage concept, this is one of many initiatives we regularly evaluate, and plans evolve significantly as we learn.”

Business Insider reported that a key technology behind the effort could come from Boxbot, an Alameda, Calif.-based robotics startup that uses conveyors and AI-driven storage trays to automatically sequence packages for vehicle loading. The company says the process is up to 10 times faster than manual methods.

Boxbot has raised $29.5 million from investors including Toyota Ventures, Playground Global, and Maersk Growth.

Responding to an inquiry from GeekWire, Boxbot CEO Austin Oehlerking said he could not comment on activities with any specific customer but said the company has “tested and deployed live systems within the parcel delivery, logistics, and automotive industries over the last several years.”

Oehlerking described Boxbot’s technology as filling a gap in warehouse automation. Automated storage and retrieval systems are typically designed for fulfillment operations, while Boxbot is building them for high-throughput package handling at other points in the supply chain.

“This type of storage system can be very useful at different points in the supply chain, depending on the customer,” he added.

Amazon said its delivery station initiatives are “designed to complement and empower our workforce.” The company has been ramping up automation across its operations, with more than a million robots now deployed in its fulfillment network and plans to more than double its fleet of robotic arms this year, citing goals to improve safety, ergonomics and efficiency.

The company has also opened a Last Mile Innovation Center in Germany, where it has been testing delivery station technologies including automated unloading, sorting, and scanning systems.

More than 20 states have now passed the “Washington model” of privacy legislation. Washington state hasn’t. 

In the years since then-state Sen. Reuven Carlyle introduced the Washington State Privacy Act in 2019, the blueprint has been adopted across the country, mandating that companies get the consent of consumers before collecting sensitive personal data, and providing consumers with the right to correct and delete their details in those databases.

In its home state, the bill stalled in negotiations between the House and Senate two years in a row. Every year since, a comprehensive privacy bill has been introduced in the Washington state Legislature but has failed to pass. 

Washington state Attorney General Nick Brown released his office’s first data privacy report Aug. 14, calling on lawmakers to pass a privacy law that would limit how much personal information companies can collect and keep in the first place.

But that proposal will face the same hurdle that has blocked efforts to pass a state privacy law for seven years: a fight over whether consumers should be able to sue companies that violate it.

“The attorney general supports greater data privacy protections for Washingtonians,” said Mike Faulk, a spokesperson for the AG’s office. “In our experience, this has proven to be a difficult subject for the Legislature to build consensus on.”

Experts say the stakes are rising as AI systems train on personal data that often falls outside Washington’s existing privacy protections. Without a baseline privacy law, they say, lawmakers also have less to build on when they try to regulate AI itself. 

Rethinking privacy

AI has rendered some parts of the Washington model moot, while making others more necessary than ever, according to policy experts. 

As states have begun to pass the first AI regulations, one of the highest priorities has been the regulation of AI-based high-risk decisions.

In Washington, for example, the state Legislature passed the Prior Authorization Transparency Act, which bars health insurers from using AI as the only basis to deny, delay or modify care. Washington state lawmakers also considered a bill to regulate the use of AI to make decisions of financial, educational, or legal consequence.

This is proving to be a much easier lift in states that passed the “Washington model,” often years before the current AI craze. That’s because Carlyle’s bill happened to include what’s now known as an automated decision-making technology (ADMT) opt-out clause, which granted residents the right to opt out of automated profiling when used for “legal or similarly significant effects.” 

Algorithmic wage and price determinations, as well as AI-based healthcare and employment technologies, could be regulated under the pre-existing privacy act, or by tweaking those laws.

“The states that have passed automated decision making laws have done so on top of existing privacy laws,” said Cobun Zweifel-Keegan, a managing director at the International Association of Privacy Professionals (IAPP). “There’s already restrictions, or at least the beginnings of restrictions, on automated decision making baked into these privacy laws. It’s a natural model to build on top of.”

Meanwhile, AI has made it more dangerous to go without a privacy law, because an absence of privacy legislation means more personal data online for AI models to access, said Kara Williams, counsel at the Electronic Privacy Information Center.

Williams said data minimization could prevent or limit companies from repurposing personal data to train AI systems. 

“It goes back to using the data for the purpose you collected it for,” Williams said. “Almost all of the data that companies have used to train AI systems or develop the algorithms that led to this moment were not collected for the purpose of training AI systems.”

Data minimization requires companies to restrict the collection and use of customer data to the service the customer requested. That often precludes secondary uses like selling it to a data broker.

The Washington attorney general’s privacy report also endorsed a data minimization standard, which the original Washington model does not include.

Carlyle said he might have written one in, if he were drafting the bill today.

“We live in an AI world with a giant vacuum in the sky, sucking up every ounce of data that exists on a person,” Carlyle said. “So I think the concept [of data minimization] makes some sense.” 

Meanwhile, experts say AI makes some elements of the Washington model irrelevant. 

Zweifel-Keegan of IAPP said those elements include the right to control, correct, and delete personal data, which was the bread and butter of Carlyle’s bill. Because LLMs are a weighted map of associated words, there is no straightforward way to selectively delete or change information once a model has been trained.  

“That’s just fundamentally how LLMs work. They’re not a table where you can go to my name and see all the other records that are associated with me,” Zweifel-Keegan said. “You can’t go in and selectively delete information.”

While states around the country that have passed the Washington model are now seeking to revise its provisions to meet the AI moment, Washington state has no comprehensive privacy law to start with.

“AI is making us rethink some of our foundational expectations of what a privacy law does,” Zweifel-Keegan said. “Washington could be the place where that happens.”

The story of the “Washington model”

In 2019, when now-retired State Sen. Carlyle introduced the Washington State Privacy Act, it passed the Senate 46-1 before dying in the House. One year later, it passed both chambers but died after a long and heated fight in conference.

Some say the bill didn’t deserve to pass after being “rewritten” by tech lobbyists. Others say the lawmakers who opposed the bill let the perfect be the enemy of the good. 

The original bill was based on an opt-out framework, also called “notice and consent,” which required a platform to present a privacy policy to users who consent to the collection of their data by continuing to use the platform. The bill’s sole enforcement mechanism was the state attorney general, and did not offer a private right of action for individuals to sue companies that violated the proposed rules. 

In 2019, Carlyle was focused on establishing a baseline notion of consumer rights — one that could be revised later, as other states ultimately did.

“At that time we didn’t have a direct understanding that consumers have a right to correct or delete their personal data, we didn’t have an understanding of what opt out meant for advertising, or an understanding of data brokers and the role that they play,” Carlyle said.  

His bill also established special protections for sensitive data and frameworks to hold corporations accountable for complying with transparency and disclosure requirements. 

“Those were pretty novel pillars that didn’t exist,” Carlyle said. “That’s why it had a big effect on other state laws.” 

By March 2021, Virginia had passed a privacy law closely modeled off of Carlyle’s template, and over the next few years, more than 20 other states did, too.

In Washington, meanwhile, no progress was made. After Microsoft endorsed the Senate bill in 2019, consumer advocacy groups and some state lawmakers said that the tech lobby’s influence had gone too far. The state House countered with a stronger privacy bill, premised on opt-in data collection frameworks and enforced by a private right of action.

Both the 2019 and 2020 legislative sessions ended in failed negotiations between the state Senate and House over their competing privacy laws. Every year since 2021, Rep. Shelley Kloba has introduced a bill that preserves the House’s stronger language. It has yet to make it to the House floor. 

A potential compromise

The sticking point for Washington negotiators in 2019 and 2020 was the enforcement mechanism. Carlyle’s bill proposed state attorney general enforcement, while the House bill, led primarily by then-Rep. Zack Hudgins, included an additional private right of action.  

Consumer advocacy groups are firm in their support for a private right of action as part of a data privacy law. 

“Attorney general enforcement alone is not sufficient to enforce privacy laws, just because of limited resources and staff and funding that attorneys general across the country face,” said Williams, the EPIC counsel. “We need a stronger enforcement mechanism, like a private right of action, that would allow consumers to vindicate their own privacy rights and to take companies to court who have violated their privacy rights.” 

For some in the tech industry, a private right of action is seen as unnecessarily harsh, stymieing innovation while AG enforcement would have sufficiently guaranteed compliance. 

“I believe that the difference is, are you looking to get companies to comply and have clear enforcement or are you looking to punish?” said Rose Feliciano, TechNet executive director of policy for the Northwest United States. TechNet is a trade association that includes tech industry giants such as Amazon and Google.

Carlyle agreed, saying his efforts failed because the trial attorneys “were not enthusiastic about giving up a right of private action against big tech.” The insistence on letting individuals sue, he said, is a case of “perfect is the enemy of the good.” 

“It’s the ultimate representation of, ‘we can’t have any regulation, any policy framework, any guidelines, any protections whatsoever, unless it’s a grand slam home run for individual lawsuits,'” he said.

The private right of action has continued to hold up privacy legislation.

Rep. Kloba’s alternative, the People’s Privacy Act, ties enforcement to the state’s Consumer Protection Act, under which a plaintiff’s private action can seek damages, attorney’s fees, and treble damages capped at $25,000. Her bill treats all violations, including failure to comply with records keeping and timely responses to consumer queries, with the same severity.

This winter, Kloba may be open to changing that. She said she’s willing to consider separating enforcement rules so that some violations would be eligible for a private right of action and others would be subject to civil penalties enforced by the attorney general’s office. 

“Over the last eight years, various laws have been put in place in different states and we’ve seen them then go back and improve them over time,” she said, “and so I think it’s time to have that conversation.”

Jargon stinks.  What do the terms open weights, RAG, and agent mean exactly? Here’s a plain English, slightly snarky glossary of befuddling AI terminology with references for further reading.

AI is a broad name for the technology. Machine learning is the part where a system learns from data instead of following rules somebody wrote, a neural network is the structure that does the learning, and deep learning just means a neural network with a lot of layers.

Here’s the nitty-gritty: the terms that get used loosely, and the distinctions the loose usage hides.

1. Model, LLM, frontier model

ChatGPT is the app you open; an LLM, or large language model, is the AI running inside it.

“Frontier” isn’t a technical category at all. It means the handful of biggest and most capable models at any given moment, so the trophy keeps changing hands.

Everyone says “LLM” and hardly anyone could define it on the spot. “Frontier model” is worse. It’s a ranking, announced by the people being ranked.

Further reading: How ChatGPT Works: A Non-Technical Primer (MIT Sloan). Rama Ramakrishnan walks through the predict-the-next-word mechanism everything else is built on.

2. Prompts, tokens, parameters

A prompt is the thought, question, or instructions you provide to the LLM (plus whatever the app added before it without telling you). The LLM takes the prompt and generates words, both in its internal “thinking” process and in the answer it shows you.  

Tokens are (roughly) the words going in and coming out. The model chops your prompt into tokens, then produces more of them as it answers, and they’re what the industry charges by.

Parameters, also called weights, are the numbers inside the model. A frontier model has hundreds of billions of them and the biggest now run to trillions, and nobody can tell you what any single one does.

Parameter counts get quoted like horsepower. The number nobody advertises is how many tokens it takes to answer your question, and that’s the one that shows up on the bill.

Further reading: The only AI glossary you’ll need this year (TechCrunch, July 2026). Its entries on tokens and weights are the clearest short treatment of the building blocks.

3. Pre-training, post-training, fine-tuning

Pre-training is feeding the model most of the internet, so it learns to predict the next word in a sentence. That’s the expensive part, and it produces something that knows a great deal but can’t follow an instruction.

Post-training is where people rank its answers and it learns to give more of what ranked well. Fine-tuning is post-training done by you, to somebody else’s model, on your data.

Pre-training costs hundreds of millions and gets you a model that won’t answer a question well. Post-training is what gets you the product.

Further reading: Illustrating Reinforcement Learning from Human Feedback (RLHF) (Hugging Face, 2022). The clearest walk-through of how ranking a model’s answers becomes a signal for training.

4. Training from scratch vs. distillation

From scratch, you buy (or rent) the computers and do the work to build and train a model. Distillation trains a cheap model on an expensive model’s outputs, so it inherits the behavior without the bill. Distillation is against most AI companies’ terms of service.

OpenAI accused DeepSeek of distilling its models, which is a bold position for a company that trained on the whole internet without asking. Learning from other people’s work is fine right up until the other people are you.

Further reading: OpenAI accuses DeepSeek of “free-riding” on American R&D (Rest of World, February 2026). OpenAI’s memo to Congress, and an analyst’s reply that no model is an island.

5. Training vs. inference

Training is how you build a model. Inference is what happens every time it answers: the model runs and produces a result.

Training is a one-time cost. Inference is a cost you’ll pay forever. Training runs for months and costs hundreds of millions; one inference, meaning one answer, costs a fraction of a cent, and it happens billions of times a day.

Training costs get announced. Inference costs get discovered. Only one of them shows up in a press release.

Further reading: Why AI’s next phase will likely demand more computational power, not less (Deloitte, 2025). Inference reaches about two-thirds of all AI compute in 2026, up from a third in 2023.

6. Open weights, open source, API-only

We typically use LLMs by accessing an app like ChatGPT, Claude, or Gemini. But experts often want the model itself, not just an app wrapped around it. Open weights means that an AI expert can download the model and run it on a server. You don’t get the data or the code that made it.

Open source means data and software that experts can use and modify, which almost no major model offers (AI2’s Olmo is a rare exception).

API-only means you can’t have the model at all. You send your text to the company’s computers, the answer comes back, and you pay for every use, which is also what’s happening when you use ChatGPT or Claude through an ordinary account.

Open weights is how you claim the open-source mantle without giving much away. Open washing, basically.

Further reading: Open-Weight Models Aren’t Enough. We Need Truly Open Source AI Models for Science and Society. (Stanford HAI, August 2026). James Landay’s term for downloadable weights without the data or code is “open distribution.”

7. Context window, memory, RAG

The context window is how much text the model can hold in mind at once, including your question and everything pasted into the conversation.

Memory is a feature that saves facts about you and slips them back into the context window later.

RAG, short for retrieval-augmented generation, searches a document collection and drops the relevant passages into the context window before the model answers.

Nothing in the model remembers you. The app keeps a file on you and pastes it in before every conversation, and that’s a less charming way to describe the same feature.

Further reading: Glossary of Terms: Generative AI Basics (MIT Sloan Teaching & Learning Technologies). Defines context window and RAG in plain language, and is careful to put the model’s “memory” in quotation marks.

8. Chatbot, workflow, agent

A chatbot answers and stops. A workflow runs the steps you defined, in your order. An agent receives a goal instead of steps, and works out for itself what to do, calling out to other software and checking the results until it’s done or stuck.

Ask about a delayed flight and a chatbot quotes you the policy; a workflow uploads the refund form you built; an agent rebooks you.

Useful test: if it decides its own next step, it’s an agent. If you decided the steps, it’s a workflow.

Further reading: Building effective agents (Anthropic, December 2024). The source of the distinction: workflows run predefined code paths, agents direct their own.

9. Hallucination, AI slop, AI cream

A hallucination is a confident falsehood, like a citation to a paper that doesn’t exist. The model isn’t lying; it has no notion of truth to violate. It’s producing text that looks like the right kind of answer.

AI slop is a different failure: accurate, fluent, and worthless. Think of the LinkedIn post that says nothing in 300 fluent words.

AI cream is the third case and the rare one: superb writing authored with the help of AI.

Nobody sets out to make slop. Everyone believes they’re making cream.

Further reading:  2025 Word of the Year: Slop (Merriam-Webster, December 2025). The dictionary definition turns on quantity: low-quality content “produced usually in quantity” by AI.

Why language models hallucinate (OpenAI, September 2025). Argues that hallucinations persist because benchmarks score accuracy alone, so guessing beats admitting ignorance.

10. Alignment, guardrails, censorship

Alignment is the research problem of getting a model to do what people want when nobody’s watching. Guardrails are the rules behind its refusals: “no, I won’t tell you how to make a bio weapon.” Censorship is a guardrail that blocked something you wanted.

The same refusal is “safety” in the press release, “guardrails” in the documentation, and “censorship” on X.

Further reading:  Model Spec (OpenAI, updated December 2025). A published rulebook for what one model will and won’t do, which makes refusals arguable rather than mysterious.

I snuck in one novel term that’s been sorely absent from the field.  Can you tell which one?

Further reading: other glossaries

Five general AI glossaries, listed roughly from most opinionated to most technical.

The only AI glossary you’ll need this year (TechCrunch). About 30 entries, written for readers who follow the industry news. Strongest on distillation and compute.

Artificial intelligence glossary: 60+ terms to know (TechTarget). The broadest of the mainstream lists, and the only one that bothers to define model collapse.

Glossary of Terms: Generative AI Basics (MIT Sloan Teaching & Learning Technologies). Twenty-odd entries aimed at people who use the tools rather than build them.

Glossary of Terms for Artificial Intelligence (Columbia Business School). The shortest and plainest. Useful as a test of which terms are unavoidable.

Machine Learning Glossary (Google for Developers). Hundreds of technical entries, and the only glossary here that defines “AI slop” a few lines away from several hundred pieces of real math.

Starcloud says it has raised $250 million in new funding to support the creation of a constellation of data center satellites powered by Nvidia’s next-generation AI chips.

The Series A extension funding round was led by Manhattan West, with participation from existing investors including Benchmark, EQT, Soma, NFX and 776. Among the new investors joining for this round are Nvidia, Cisco Investments, Cedar Capital, Goanna Capital and Standard Capital.

Founded in 2024, Starcloud is headquartered in Redmond, Wash., and is building production lines for its Starcloud-3 spacecraft at a new 100,000-square-foot manufacturing facility in Woodinville, Wash. The newly announced round brings the startup’s total capital raised to $450 million, with a post-money valuation of $2.3 billion.

Nvidia’s participation in the funding round brings Starcloud’s collaboration with the computer-chip titan to a new level. In November 2025, Starcloud flew Nvidia’s H100 GPU to orbit for the first time. It used the chip to train a large language model called NanoGPT — marking a milestone in space-based AI data processing.

Starcloud plans to equip future satellites with Nvidia’s Space-1 Vera Rubin Module, which Nvidia says will deliver 25 times as much in-space compute capability as the H100. Starcloud’s satellites will serve as an early flight platform for the space-rated chips.

“This fresh capital empowers us to build the infrastructure to launch many more of Nvidia’s most advanced GPUs into space,” Starcloud co-founder and CEO Philip Johnston said today in a news release.

Starcloud says the new investment will fund the continued buildout of manufacturing capacity, engineering work in collaboration with Nvidia and the procurement of future launch slots. Manhattan West’s Lauren Selig will join Starcloud’s team as a board observer.

Starcloud has filed an application with the Federal Communications Commission to operate as many as 88,000 satellites as orbital data centers for AI and other applications. It’s not the only company targeting the market for orbital data centers. Most notably, SpaceX has filed its own plans to put up to a million data center satellites in space, for a project called Starmind.

The push to move AI infrastructure into space is driven by growing terrestrial bottlenecks surrounding land, power and water consumption — and by the political controversies those bottlenecks have sparked.

BELLEVUE, Wash. — There is no prominent signage outside Anduril’s downtown Bellevue office. The name isn’t listed in the building directory, on the suite door, or in the lobby. That may be by design: Inside, Anduril engineers are quietly building AI-driven, autonomous technologies intended to keep the U.S. and its allies ahead in a rapidly changing military landscape.

Step through the front doors and it becomes clear that this is no ordinary tech company. 

Scattered across desks, production tables, conference rooms and showroom labs are autonomous drones ranging in size from a microwave to a ping-pong table, alongside next-generation digital night-vision goggles and military-grade edge computers designed to keep soldiers connected in an increasingly data-rich battlefield. 

The Costa Mesa, Calif.-based defense giant — which closed a $5 billion funding round at a $61 billion valuation earlier this year and is reportedly targeting a $100 billion valuation — is rapidly building a massive engineering footprint in the greater Seattle region. It includes modern offices in downtown Bellevue and Seattle, and a multi-acre military testing facility near Carnation, Wash.

GeekWire got a peek inside the iconoclastic company’s Seattle-area operations this week. We toured the Bellevue offices, demoed products, and sat down with senior vice president Tom Keane. The former Microsoft cloud executive leads Anduril’s local presence and oversees its connected warfare division, including mixed reality, AI and edge computing technologies.

Keane was clear about the company’s growth ambitions in the Seattle region: “I could absolutely see a world where we get to 1,000 people here in terms of engineers,” he said.

Asked when that milestone could be reached, Keane said, “It’s sort of almost as fast as you can hire people.” Realistically, that will likely occur over the next two years.

Anduril currently employs 560 people in its Bellevue and downtown Seattle offices, up from about 30 four years ago. That’s when Keane, who had interviewed with 62 companies looking for a new challenge in edge computing, joined Anduril because of what he called “world-class teams in every discipline.” 

With more than 110 open positions in Washington state and a newly leased third floor in downtown Seattle coming online, Keane said the regional expansion is only accelerating.

The regional workforce was until recently slated to include a maritime manufacturing hub along the Lake Washington Ship Canal in Seattle. However, as GeekWire reported this morning, Anduril has since vacated the former Foss Maritime shipyard space after the U.S. Navy canceled the autonomous warship program the company was pursuing.

But Anduril’s maritime ambitions haven’t gone away. Alongside its autonomous aircraft, low-cost cruise missiles and counter-drone interceptors, the company continues to develop a large subsea fleet, from modular survey vehicles to the extra-large, school-bus-sized Dive-XL.

The Big Tech pipeline meets ‘grit’

The company’s Seattle-area ramp-up is a play for the region’s engineering talent — a mix of disciplines well-suited to a next-generation defense hardware company, and one Keane says is hard to find in one place. That includes distributed systems engineers, AI and machine learning specialists, and advanced optics experts.

Another advantage: proximity to major military installations including Joint Base Lewis-McChord and Naval Base Kitsap.

Tech giants such as Microsoft, Amazon, Google and Meta have long dominated local engineering hiring. But Anduril is recruiting aggressively from those companies, pitching engineers who want to move fast and see their work in the field.

Deep pockets and close ties to the U.S. defense establishment don’t hurt, either.

Anduril got a large infusion of that talent 18 months ago, when it took over Microsoft’s IVAS (Integrated Visual Augmentation System) program along with about 100 local engineers who had spent years developing augmented reality hardware for the U.S. Army.

“If you take machine learning and AI… you can look out any of these windows and see places that we can recruit from,” Keane said from a conference room overlooking downtown Bellevue. “For a lot of software engineers, this is incredibly applied.”

Keane acknowledges the company’s culture isn’t for everyone. In fact, Anduril got plenty of buzz last year with its edgy recruiting campaign, “Don’t Work at Anduril.“

Rather than plush tech perks and work-from-home flexibility, Anduril pitches what Keane calls “grit” — a willingness to take ownership of hard engineering problems and test them in the field. It’s very much an in-office culture, in part because the company builds hardware.

Anduril also moves fast, an approach Hawaiian-shirt-wearing founder Palmer Luckey wields against slower traditional military contractors.

The company’s “grit” translates directly into how Anduril tests its gear.

In rural Carnation, Wash., about 20 miles east of Bellevue, Anduril operates a testing range with a built-out “shoot house” for low-light tactical work. It gives engineers a place to iterate on code and hardware in the mud and rain alongside active-duty warfighters.

Among them: members of the Army’s 75th Ranger Regiment from nearby Joint Base Lewis-McChord, who recently put Anduril’s EagleEye mixed-reality system through its paces on the firing range and obstacle course.

Hands-On with ‘EagleEye’ and Lattice

During a walkthrough of the Bellevue facility, Anduril’s team showed us a wide range of hardware being designed, tested, and calibrated on site.

The centerpiece was EagleEye, the company’s heads-up hardware and software suite built for warfighters. Historically, soldiers have had to lug 3 to 5 pounds of glass and electronics on their helmets, paired with fragmented radios, specialized batteries and thick cables.

Anduril’s system strips that down to a 115-gram pair of augmented reality glasses — about four ounces — built in collaboration with Meta (for waveguides), Canon (for sensors), Qualcomm (for custom silicon) and Oakley (for ballistic protection).

I tried on the lightweight camouflage backpack and glasses — no helmet in this case — in a demo in the Bellevue testing lab. It ran through three operational “vignettes” driven by Anduril’s core Lattice software:

• Tactical HUD: A heads-up display projecting friendly force markers, compass headings, and mini-maps into the user’s field of view so troops don’t have to look down at a handheld screen while moving.
• Threat Detection: Real-time alerts from external AI sensors — such as a tower camera or an overhead drone — highlighting incoming aerial threats directly in the display.
• Command & Control (C2): The ability to assign tasks to autonomous technology. Using a small handheld controller, an operator can send a virtual Ghost drone to a location, pull up a live video feed and execute a simulated strike.

In each case, the user controls the system — navigating between maps and text — with a click of a button on a small device connected to the front of the backpack.

The team also previewed its upcoming digital night vision system. Instead of the narrow “toilet paper roll” view of traditional analog goggles, Anduril’s VR-style digital displays offer an 84-degree field of view. That’s more than double the 40 degrees warfighters get in devices today.

Using machine learning models, the system fuses thermal imagery and low-light camera feeds in real time, making heat signatures leap out in pitch-black environments.

To ensure every pixel lands accurately without causing motion sickness, Anduril’s engineers use custom robotic arms to run geometric calibration on each lens distortion map before it leaves the facility. An engineer running the testing equipment politely asked GeekWire not to photograph the system.

The night vision system has not yet been deployed but will be part of the Soldier Borne Mission Command prototype delivery to the U.S. Army next year, said Stephanie Davis, communications manager for Anduril’s connected warfare group.

Protests, hiring and hardware

Operating a high-profile defense company in the Seattle area doesn’t come without friction. Activist groups recently staged protests outside Anduril’s downtown Seattle offices, targeting the company’s autonomous weapons development and military contracts.

At the time, Anduril issued a statement to GeekWire saying that it respects the right to free speech.  “That said, it is perplexing when people choose to protest a company dedicated to supporting the very military that safeguards those rights,” the company added.

When asked about the pushback, Keane said critics don’t affect Anduril’s hiring. “There’s protests at every company, frankly.”

Anduril’s hiring reaches well beyond Big Tech. Drew Swanson, a lead architect, spent 13 years in the military working in communications supporting special operations units in the field. He described an Anduril technology called Squad-A, which helps squad leaders pick the best network connections to stitch together incoming intelligence feeds.

Swanson said the draw of Anduril is the chance to escape corporate bureaucracy and tackle high-stakes physical problems, reflected in comments he gets from others in Seattle tech.

“One of the biggest things that I always hear is: ‘I want to come over there because you guys are going after really hard problems,'” Swanson said of conversations with tech colleagues around Seattle.

Solving those problems often comes down to fundamental hardware redesigns, like combining a soldier’s ballistic vest, battery and edge computing device into a single body plate. Keane showed off a prototype as the tour wrapped up — an example of Anduril’s broader approach.

“Batteries are such a huge thing,” he said, pointing to the body armor. “If you put the three of them together… you can start to get rid of stuff. You take what was historically three separate heavy things and turn it into one software-enabled platform.”

The Seattle region outranked New York, Austin, Boston and other tech hubs, trailing only the Bay Area, in an annual tech talent scorecard from commercial real estate firm CBRE that weighs factors such as tech worker concentration, wages, education levels and real estate costs.

You may have seen headlines this week that New York overtook the Bay Area for the first time in the CBRE rankings. That was based on a subset of the data: a straight head count in each market. New York’s 394,300 tech workers topped the Bay Area’s 375,730. Seattle ranks seventh on that specific list, with 213,010 tech workers across the region.

But in the broader scorecard, Seattle held onto the No. 2 spot (which it also occupied last year), thanks to the density of its tech workforce, one of the largest concentrations of AI talent in North America, and the second-highest tech wages on the continent.

The market-by-market workforce figures in the report run through 2025, so this year’s layoffs aren’t reflected in the rankings. CBRE does flag the trend nationally: the tech industry accounted for a record 31% of all U.S. job cuts through June, up from 13% for all of last year.

Some of the Seattle region’s specific strengths:

The AI workforce is deep. Seattle is home to 41,591 workers with AI skills, third most in North America, behind the Bay Area and New York. One in five of the region’s tech workers now has AI skills — a higher share than anywhere except the Bay Area.

Tech is a bigger part of the economy here. Tech jobs make up 10.2% of all employment in the metro area, among the top five markets and nearly double the 5.5% average across the 50 markets studied in the CBRE report.

Wages are in a tier of their own. Seattle’s average wage for tech workers at tech companies was $190,050 in 2024, second to the Bay Area’s $211,048, and nearly $50,000 above third-place Boston.

The workforce grew while the Bay Area’s shrank. Seattle added 24,590 tech jobs from 2022 to 2025, a 13.1% increase and the fifth-largest gain of any market. The Bay Area lost 23,900 jobs over the same time period.

However, the report also points to warning signs:

Many offices are sitting empty. The Seattle metro area’s office vacancy rate hit 28.6% in the fourth quarter of 2025 — the highest of the 50 markets in the report. That’s despite 1.9 million square feet leased by AI companies across the region since 2023, according to CBRE.

Costs are near the top. Seattle is the third-most-expensive place to run a 500-person tech company, at $73.9 million a year in wages and office rent, behind the Bay Area at $90.6 million and slightly behind New York, which edged Seattle by about $24,000.

Young workers are going elsewhere. Seattle’s 20-something population fell between 2019 and 2024, even as its share of 30-somethings grew to the highest of any market in the report. The region is drawing mid-career but not entry-level talent, which risks creating a thinner pipeline over time.

One counterweight to the pipeline concern: the University of Washington ranks fifth among U.S. universities for its AI program, according to CBRE’s analysis of U.S. News & World Report rankings — the only school outside the Bay Area, Boston and Pittsburgh in the top five.

Access the full CBRE Scoring Tech Talent 2026 report here.