© AP Photo/Alex Brandon
© AP Photo/Nam Y. Huh
© Getty Images/KanawatTH
© Amelia Brust/Federal News Network
Summary Bullets:
• As telemetry arrives, Observe adds structure that reflects each customer’s business, creating a context graph, which gives an AI agent a roadmap for querying data.
• Querying a structured database is quicker and burns far fewer tokens, which helps address the AI cost concerns that are top of mind among enterprises.
The need to manage costs and recognize a positive ROI are top concerns among organizations rolling out generative and agentic AI. Even the largest enterprises are taking a more careful look at token economics as they seek to scale the use of AI tools throughout their organizations. As employees, developers and agents all gain access to generative AI and use an ever-growing number of tokens, spending is skyrocketing. During the Snowflake World Tour 2026 in New York City in October, Jeremy Burton, former CEO of Observe and currently General Manager of Snowflake’s Observability business, notes that companies let chaos reign while they experimented with AI and now need to rein in token usage. One often-overlooked lever is context. The context graph helps agents investigate operational telemetry more efficiently, which can improve answers and reduce token usage in those workflows. For this reason, the impact of Snowflake’s acquisition of Observe, which helps organizations better structure and understand how data is being used, has the potential to manage overall enterprise AI spending as well.
Snowflake’s acquisition of Observe expands its observability portfolio. As telemetry arrives, Observe adds structure that reflects each customer’s business, creating a context graph that maps semantics and relationships across logs and metrics, which gives an AI agent a roadmap for querying data. An agent could dig through raw data, but querying a structured database is quicker and burns far fewer tokens. As Observe puts it, the company does not have better AI; it has better data. Observe ingests data based on OpenTelemetry open standards and can store it in Apache Iceberg formats, providing greater flexibility and control. It uses object storage, elastic compute, and interoperable standards to meet customers’ requests for flexibility and cost sensitive solutions.
Observe complements Snowflake Cortex AI Gateway, announced in July 2026. The gateway governs how Snowflake and third-party agents access tools, models and data. It routes requests to approved models based on cost, quality, and latency; allocates spend to teams and workloads; sets limits; and records agent actions. Observe Agent Observability goes deeper, tracking which tools agents access and correlating agent activity with backend telemetry. It adds more detail directly from the agents themselves, such as connections to underlying application and infrastructure behavior. Observe stores the telemetry in Snowflake on low-cost object storage. At the scale of telemetry data, scan-based queries using Observe’s architecture are cheaper and faster than using index-based tools. Together, Cortex AI Gateway and Observe provide an end-to-end picture of an agent, from chat client to gateway to back end. Plenty of tools observe the agent itself; Observe provides deeper insight into the surrounding telemetry.
Improved context sits at the center of several trends related to how data is accessed. Software is going headless, as users prefer prompts and APIs to complicated interfaces when interacting with information. Walled gardens are crumbling as APIs, CLIs, and MCP streamline access to information and enable AI agents to drive actions. Evaluations, which interrogate how an agent reached its answer in testing and production, depend on rich telemetry. The next opportunity lies in context graphs. Enterprises that structure their data first can make more efficient use of tokens. With AI bills climbing, vendors that offer solutions that reduce token usage offer a compelling value proposition.
The post Snowflake’s Observe Helps Businesses Rein in Token Usage by Improving Context appeared first on IT Connection.
© Federal News Network
© Amelia Brust/Federal News Network
© Federal News Network
© Getty Images/burcu demir
© Federal News Network
© AP Photo/Stephanie Scarbrough
© Getty Images/iStockphoto/KanawatTH
© The Associated Press
© Federal News Network
Summary Bullets:
• AI-driven conversational discovery experience boosts the relevancy of marketplaces among broader audiences.
• Network services evolve into consumable and programmable products via enterprise agents.
Agentic AI innovations have boosted hyperscaler marketplaces’ relevance, and global telecommunications giants are taking notice.
Struggling for years on trying to fit a square peg into a round hole, telcos are recognizing the online storefronts as a prime route to sales growth by leveraging agentic AI to streamline enterprise delivery of network services. Factors range from agentic AI-driven product procurement through conversational discovery of solutions, to improved productization of infrastructure services.
Advancements in open-source software (OSS) have resulted in a greater number of carrier solutions being listed on marketplaces, alongside standardized APIs (enabled through MCPs, GSMA Open Gateway, and CAMARA) for streamlined connectivity. Networks (e.g., networks-as-code) are evolving and becoming more consumable and programmable via enterprise agents, with no complex coding or integration required by enterprises. Ericsson-led network-API operator alliance, Aduna, is a good example of such activities. Nokia’s contribution of Network-as-Code APIs to Google Cloud Marketplace is another.
Telcos are filling key roles through marketplace channels, including fulfilling private wireless deals, reselling software and platform services, and offering enterprise customers valuable consulting and integration services.
This infra productization couples nicely with another trend in marketplaces: Agent mode UX innovations have appeared across marketplaces. They support conversational discovery experiences for software procurement and are significant for their role in broadening the audience of less technical buyers.
Within the last year, AWS released two AI-powered tools for its marketplace: agent mode, which is its version of a conversational discovery experience for software procurement; and enhanced search. Buyers describe to the agent a use case, including specific requirements, to gather product details and compare options. These conversational queries in marketplaces are significant to sellers for their role in expanding the audience to include less technical buyers looking to solve their companies’ digitization issues.
Google and Microsoft also released chat-style methods to discover appropriate solutions on marketplaces. Google’s AI-powered search experience moves beyond keyword searches to provide users with natural-language discovery across its services and AI agent catalog. Its approach is similar to AWS by allowing users to describe their goals or a technical problem in order to better identify appropriate products, including third-party offerings. Microsoft brands its conversational AI experience as ‘intelligent discovery’, with interactive chat-enabled evaluation and comparisons of products.
A lot is at stake in the battle of the hyperscaler marketplaces, reported to be worth tens of billions of dollars, including partner solutions. AWS now claims over 30,000 product listings, including an AI Agents and Tools category launched last July and later that year, and agent mode for AI-enhanced search for buyers. Microsoft rebuilt its marketplace last September after merging its Azure Marketplace and AppSource. The storefront now includes over 3,000 AI apps and agents. Adding AI Agent Marketplace last year, Google differentiates on a cost basis, cutting its fee from a flat 3% down to 1.5% for certain deals. The marketplace wars will heat up as the telcos increasingly play in this space.
The post Agentic AI Transforms the Marketplace UX as Infra Services Become Consumable Products appeared first on IT Connection.
Summary Bullets:
• Technology vendors are investing in forward deployed engineers to adapt AI solutions to meet customer needs.
• Systems integrators are working alongside their technology partners to help enterprises accelerate their AI adoption and ensure business outcomes are met.
It is widely reported that only a small percentage of firms that have conducted AI pilots are moving them into production. Even fewer enterprises have been able to scale their AI initiatives across multiple business functions. Based on GlobalData’s observations and discussions with enterprises, there has been greater success in AI implementations among large enterprises that have engaged a systems integrator or consulting firm to work alongside their technology partners, including hyperscale cloud providers and AI model companies (e.g., Anthropic and OpenAI).
This is not surprising since generative AI (GenAI) and agentic AI are relatively new and rapidly evolving. There are unknowns for enterprises in harnessing the technology, both from the technical and commercial standpoints. Moreover, the use of AI is tied to business objectives, not an IT-led initiative. This suggests a need for a high-touch, consulting-led approach in which technology companies work closely with enterprises to turn AI into solutions that deliver against business outcomes. Systems integrators and consulting firms do not just provide the technical knowhow, but also business consulting capabilities and industry knowledge.
While AI studios and labs are useful in demonstrating the possibilities, actual deployment can be challenging for large-scale enterprises that have complex tech stacks with legacy systems, and messy workflows. Increasingly, technology vendors are embedding forward deployed engineers (FDEs) into client premises to support customer teams with their AI implementations. Technology vendors (e.g., AWS, Databricks, Google Cloud, Microsoft, and Salesforce) have FDEs that are highly skilled in their technology to help customers solve problems and develop custom solutions (e.g., sorting out data pipelines, configuring data models, and setting up guardrails and security controls). However, it is harder for software vendors and to some extent the hyperscalers to maintain large FDE workforces. They are therefore working with service provider partners to develop FDE teams to jointly work with clients.
For systems integrators, the goal is not just to help clients move AI into production: It is also to build long-term relationships with clients, and they are more focused on helping clients through their AI lifecycle. Cognizant is responding to this trend with its Frontier-certified workforce. This comprises two new job categories: Frontier Certified Engineers (FCEs) and Frontier Business Operators (FBOs). FCEs focus on reimagining work processes and identifying where AI can make the most business impact. They work with clients directly to realign business processes for an AI-enabled environment. On the other hand, FBOs take full ownership of the operational results via the realignment of business processes. They manage a workforce of humans, AI agents, and automated software to ensure tasks are assigned to the most efficient resources and expected business outcomes are achieved. The company plans to scale to 5,000 FCEs and 10,000 FBOs by Q4 2026.
Capgemini has introduced its outcome deployed engineer (ODE) model, aimed at supporting the agentification of clients and optimizing the time-to-value. Capgemini’s approach is to embrace technology partners’ FDE model, but focus on delivering the business outcomes clients are aiming for. This helps the company shift its engagement from a talent-driven model to a value/outcome-based model. The company is building its ODE by aligning to key technology partners. It started with Google Cloud; Capgemini’s ODEs work alongside Google Cloud FDEs in dedicated pods within a client’s environment to design, build, and deploy AI agents around workflows while committing to business outcomes. The company is expected to have similar arrangements with more technology partners going forward.
Each systems integrator has or will have a similar FDE-like approach to engage customers in AI projects focusing on business outcomes. The aim is to help customers achieve faster implementation and return on investment (ROI). After having AI in production, enterprises also have to grapple with the rapid advancement of frontier models, the changing pricing mechanism, and an increasing number of use cases. Systems integrators add greater value to customers by supporting their AI lifecycle with continuous enhancements, including providing the AI technology (i.e., the partner ecosystem), the necessary tools (e.g., agentic AI builders and control plane), and the human expertise (technical, business, and industry knowledge).
The post Systems Integrators Adopting the FDE Model with a Focus on Business Outcomes appeared first on IT Connection.
Summary Bullets:
• While IP voice is less ubiquitous, when it matters it, it really matters
• Agent-to-agent communication adds a new paradigm to IP voice services
Calling IP voice a legacy service is a little unfair, but only a little. Innovation in this arena is still happening as can be seen by BT’s recent launch of its automated UC edge platform designed to help enterprises move numbers between UC platforms automatically. However, a recent and stark counterpoint to this is Lumen’s decision to exit the IP voice market.
Lumen’s decision is based on multiple reasons. The provider is focusing on what it sees as its core value proposition, which is its network services capability. Further, after a period of selling-off much of its global infrastructure, the acquisition of Alkira has broadened the provider’s horizons again and it now intends to be a provider of global AI-networking solutions. Lumen felt that IP voice was not something it could offer as a global proposition and so decided to exit the market. While Lumen may have had broader strategic reasons for exiting, there is no doubt that if Lumen had seen IP voice as a high value, high growth opportunity then it would have acted differently.
Lumen is somewhat of an outlier, however, and many providers are, if not bullish, then at least moderately optimistic about the future of voice services. For example, BT International has made international IP voice one of its core service offerings and will continue to invest in it. BT reports that IP voice has been one of its biggest success stories in winning new logos. AT&T is also doubling down on voice, albeit with a focus that, while still possessing a global aspect, is America-centric.
What is powering this optimism when many enterprises have already taken IP voice services away from many of their workers in favor of mobile voice and video conferencing solutions? One part of the answer lies in customer contact services. Contact centers are viewed by many providers as a key growth market and voice is a key part of this. It is easy to assume that the growing use of AI is eroding the voice component of contact centers, but this is not necessarily the case. AI chatbots have automated many simple processes such as order tracking, password resets, and similar. However, extending AI into more complex and regulated environments is less easy. The token cost of agentic AI also makes the cost benefits less certain.
As AI continues to improve the cost comparison will change, but there is also the customer preference aspect that should be taken into account. Many customers continue to prefer talking to a human rather than engaging with AI. This is true among the younger generations as much as older generations. There is also the fact that where humans most want to speak to another human is in the most important interactions – when there has been a problem, or selling high value services or situations where customer retention and revenue generation are at stake. While there may be fewer calls into contact centers, when they happen, they matter – and their importance is creating greater interest in high quality voice services.
The emerging factor keeping service providers interested in IP voice is also centered around AI. In particular, it is calls that will be generated by agentic AI agents – either calling humans, or interacting with other agentic agents. As businesses and individuals begin to deploy more agentic AI agents they will need to communicate and it is not yet clear how this will happen. Some of it will be via APIs and MCPs, but not all systems will be connected, and agent-to-human-scenarios will always require either voice or text.
So, while AI is a clear threat to IP voice growth, it is also the potential savior of this oft-maligned but still hugely important technology.
The post Agentic AI May Give IP Voice a New Lease on Life as Legacy Decline Continues appeared first on IT Connection.
Summary Bullets:
• Meta will need to shed its consumer-centric image to gain traction in the business market, even if it only targets smaller organizations.
• Convincing organizations that they can trust Meta to secure personal information and corporate intellectual property will take time and effort.
On Monday September 28, 2026, Meta announced plans to enter what is already a crowded enterprise AI market. It will develop Meta Enterprise Platform, a collection of tools and services that will help business customers implement their own AI strategies. Few details were shared other than the platform will incorporate Muse agent, Meta Business Agent, Muse API, Muse Code, and other solutions.
In addition to plans for an enterprise AI platform, Meta revealed it has developed a version of its Muse personal assistant for enterprises. Muse for Small Business, announced one day after Meta Enterprise Platform, expands on the consumer version of the Muse AI agent announced earlier this month. It can access a variety of enterprise applications, including Asana, Box, Canva, Dropbox, Figma, Granola, HighLevel, Intuit QuickBooks, Klaviyo, Lovable, Notion, Shopify, Slack, Stripe, Zoom, and Facebook and Instagram business accounts. A limited amount of Muse usage is available for free; customers that want to consume it at higher volumes can purchase subscription plans based on token usage.
Meta will need to shed its consumer-centric image to gain traction in the business market, even if it only targets smaller organizations. Bringing onboard a strong enterprise focused leader, Chirantan Desai, as the ‘Chief Enterprise Platform Officer’ is a good first step. The industry veteran has extensive experience gained at MongoDB, Cloudflare, and ServiceNow, which are well respected enterprise-focused brands. However, he will be fighting an uphill battle as he seeks to position Meta as a leading provider of enterprise-grade solutions. The Meta (and Facebook) brands are strongly associated with consumer services and devices, further bolstered by the recent buzz around its AI glasses and Meta Quest googles, which though intriguing, have yet to see serious application by the business community.
Furthermore, privacy concerns abound regarding Meta/Facebook. The company built a business model based upon the sharing of information, which is the antithesis of a corporate approach to data management. Convincing organizations of all sizes that Meta has the pedigree and experience to offer secure solutions to which they can entrust the personal information of customers and employees, as well as corporate intellectual property, will take time and effort.
Additionally, Meta is seeking to enter a very crowded market. All major software vendors, such as Oracle, ServiceNow, Microsoft, and Salesforce offer platforms on which their customers can create AI agents. And most claim to integrate with outside data sources. The large hyperscale cloud providers are already major players in the enterprise AI platform space, with Google and AWS positioned as leaders in GlobalData’s latest Cloud-based Enterprise AI Platform Competitive Landscape Assessment report.
Out-of-the-box solutions that can be customized are only a fraction of what is required for a successful enterprise AI strategy. Leaders in the market for enterprise AI platforms need to demonstrate they have the tools to implement responsible AI, including guardrails and harnesses for agentic AI. They will also need to offer observability and orchestration across agents and language models, as well as integration with predictive AI. And as implementations become more complex, the ability to offer professional services, whether in-house or in partnership with consultants and systems integrators, will be critical to helping customers deploy a strategy that focuses on value and business outcomes.
The post Meta Will Need to Shed its Consumer-centric Image to Gain Ground in Enterprise AI appeared first on IT Connection.
Summary Bullets:
• Three voice-activated features coming to Google Workspace reflect a broader trend of AI becoming incorporated into everyday workflows.
• The features also mirror how vendors are opening their team collaboration platforms to support the multi-platform environment in which their users operate.
Google recently announced three voice-activated AI features coming soon to Google Workspace. ‘Gmail Live’ lets users ask conversational questions to retrieve information from their Gmail inbox such as “What’s my flight’s gate number?”. Users can talk to ‘Keep Live’ to capture ideas that cross their mind and turn them into organized, actionable notes. With ‘Docs Live’ users can engage in real-time conversation to create documents on the fly. Although Workspace users cannot avail themselves of the features quite yet, ‘Gmail Live’ and ‘Keep Live’ are available to Google AI subscribers on the Plus, Pro, and Ultra plans while ‘Docs Live’ is reserved for Pro and Ultra subscribers.
Google, like all vendors in the team collaboration space, is integrating AI agents and assistants into everyday workflows. In theory this allows users to remain head-down on their work and avoid shifting gears to access AI in a separate space. The new features are intended to fulfil that vision, and they do to a degree but there is a cleaner approach.
Rather than users invoking a given feature, in the course of conducting their work users could simply describe what they would like to accomplish and AI, working in the background, would decide which feature should be leveraged to complete that goal. If the user is best served with information living in Gmail, then AI would tap Gmail Live. If a set of well-organized notes would do the trick, then AI would leverage Keep Live. If the user’s task could be best represented with a concisely written document, then AI would engage Docs Live. Regardless of the route AI chooses, everything would happen behind the scenes, transparent to the user.
In addition to reflecting how vendors are merging AI and the flow of work, the features mirror a second trend in the team collaboration space.
Rivals have acknowledged the need to open their platforms to support the multi-platform environment in which their users operate. Specifically, AI is moving from use in silos to being leveraged on a far grander scale – across vendor platforms, joining parts of organizations, and linking organizations with external partners, suppliers, and the like. AI is increasingly serving as connective tissue, threading sections of vendors’ platforms such as meetings, chat, and calling; establishing links between those platforms and third-party applications used in various parts of the business such as CRM; and integrating platforms from different vendors making things possible such as attending a meeting on one platform with an external partner that uses a competing platform.
The expanding reach of AI comes with a heavy price. Compliance, confidentiality, and security issues, among others, multiply when platforms are intra- and inter-connected with data flowing across boundaries and AI agents manipulating and disseminating that data. Vendors have made assurances regarding the integrity of their platforms, but there remains a sense that these issues have not been adequately addressed. Vendors need to take a much deeper look.
The post Google is Keeping Workspace on the Cutting Edge appeared first on IT Connection.
Summary Bullets:
Nokia didn’t use the term “Physical AI” in its announcement, but its new Cognitive Operations (CO) platform reads like a carefully constructed argument for why it belongs in that conversation. Physical AI is not just AI applied to industrial data; it is AI that perceives, decides, and supports action in the physical world, often in real time, often with incomplete connectivity, and often where failure carries operational, financial, or human consequences. This is the exact scope that Nokia is targeting with CO, initially for mining, construction, public safety, and defense. Specific examples of platform capabilities include vehicle-as-a-node, local inferencing (GPU/TPU), video/sensor analytics, safety interventions (driver fatigue/collision), and independent continuity (where operations continue despite disconnection).
The product story is straightforward and strategically on point: unify mission-critical communications, edge computing, and operational AI into a single platform that can be deployed in the field. From the point of view of its telecom heritage, the significance is that Nokia is moving up the stack without abandoning its core advantage in networks. CO is built around the idea that operational intelligence is only as reliable as the connectivity and assurance beneath it. The Cognitive Edge Node (CEN) extends this logic to vehicles and remote sites, combining multi-access connectivity (5G/LTE, Wi-Fi, satellite-ready links, LoRa, and mesh integration, where software picks the best connectivity for what is needed at the time) with local processing of data including video analytics, and AI acceleration. In effect, Nokia is selling the ability to keep operations connected and intelligent even in a harsh environment and when the cloud is unreachable.
The differentiation is not about the benefits of any particular wireless technology, rather it’s more about economics and risk. Many industrial organizations have assembled digital operations tools as a patchwork of separate asset tracking, network monitoring, video, device management, and incident coordination systems—usually from different vendors with separate interfaces and support contracts. Nokia’s promises offers a “common operational picture” instead: one operational view that fuses observability of digital twins, video analytics, predictive maintenance, and network/RF intelligence. If that promise holds true in real deployments, it will reduce downtime, speed up incident response, and limit the cost associated with a more fragmented approach.
A key element of Nokia’s “right to play” is credibility in the messy space between IT and the field. CO emphasizes deterministic behavior, SLA-driven device and traffic management, and the ability to operate when connectivity is degraded or intermittent. That’s a subtle but important argument against cloud-first industrial AI approaches that assume there will be stable backhaul connectivity and centralized processing. In mines, disaster scenes, and other contested or constrained environments, latency and reliability aren’t just “nice to have”. They are what decides whether a warning arrives in time, whether video analytics are usable, and whether communications get through when networks are congested.
But for Nokia, the “platform” narrative will only become credible when customers can point to repeatable deployments at scale. Cloud hyperscalers and industrial incumbents will argue they already deliver the AI layer, and they’ll try to make connectivity a feature, not the foundation. If Nokia can’t show rapid, production rollouts—beyond early reference pilots—its CO “platform” risks being perceived as a packaged bundle of edge hardware and dashboards rather than a new operational standard. Nokia’s decision to offer CO both on-prem (including air-gapped scenarios) and via Microsoft Azure Marketplace helps to counter that perception by giving a choice, but execution will matter when it comes to deployment timelines, integration effort, and partner-led solution breadth.
Nokia’s go-to-market packaging is notable. CO was launched with three vertical applications and a compelling concept for emergency services: “Vehicle as a Node,” where each equipped response vehicle becomes a self-organizing communications and intelligence node at the incident scene. This concept helps translate abstract edge/AI capabilities into recognizable buyer outcomes, which is how new product categories can win budget. (The other vertical applications are Cognitive Operations for Mining and Cognitive Operations for Defense.)
Physical AI will be crowded—hyperscalers, industrial automation giants, robotics firms, and private wireless vendors all want a piece of it. Nokia’s bid is to own a chunk of the operational edge layer where connectivity, compute, and assurance converge. Without explicity saying so, it’s asserting: if physical AI is going to work in the real world, it needs a network-native, field-hardened foundation. And that Nokia plans to be that foundation.
The post Without Saying So Directly, Nokia Asserts its Right to Play in Physical AI appeared first on IT Connection.
Login