❌

Normal view

There are new articles available, click to refresh the page.
Before yesterdayMain stream

Why order submitted does not mean order filled

By: Fensory
19 October 2026 at 02:24

An AI agent can call the right tool and still be wrong about the outcome. The order lifecycle starts after submission.

A trading assistant says, “Done. I bought BTC.”

What it may actually mean is, “The server accepted my request.”

Those are not the same event.

An order can be submitted, rejected, left open, partially filled, filled at several prices or cancelled before anything trades. If an AI agent collapses all of those states into “success,” the conversation stops matching the account.

That is not a language problem. It is an order-state problem.

The short answer

Submitted means the order request reached the trading system. Filled means a matching trade actually occurred.

Between those two states, the venue may validate the request, reject it, accept it into the order book, fill part of it or leave it open.

The agent should report the state returned by the account or venue. It should not infer a fill from a successful tool call.

What happens before submission

Before an order reaches a venue, the trading system may check:

  • Whether the account is authenticated
  • Whether trading is enabled
  • Whether the market is available to the account
  • Whether the order is inside an account limit
  • Whether the symbol, side, size and order type are valid
  • Whether the account has enough available balance or margin

A request can fail at any of these checks.

If an account rejects an order above its limit, no order was submitted to the venue. The useful response is a rejection that names the failed rule.

Submitted

Submitted means the request has passed the initial path far enough to receive an order identifier or acknowledgement.

It does not prove that the order traded.

A limit order away from the current market may remain open. A market order may be rejected for insufficient balance. A venue may accept the request and only fill part of the requested size.

The order identifier is the beginning of the audit trail, not the outcome.

Open

An open order is resting and waiting for a match.

The account has accepted it, but the requested quantity has not fully traded. The order may remain open, fill later or be cancelled.

This matters for an agent because open orders still affect the account. A second instruction can duplicate exposure if the agent reads positions but forgets to read outstanding orders.

Before placing another order, the agent should check both.

Partially filled

A partially filled order has traded some, but not all, of the requested quantity.

The result should include the filled amount, the remaining amount and the average fill price so far.

Calling this state “done” creates several problems. The agent may assume the full position exists, place an offsetting order for too much or leave the unfilled remainder active without telling the user.

A partial fill is an ongoing state, not a footnote.

Filled

Filled means the full requested quantity has traded.

Even then, the final result may differ from the instruction in ways the user needs to see.

A market order can fill across several prices. Fees can change the final account balance. A futures position may use margin rather than the full notional value. The report should come from the execution record, not from the model’s estimate before submission.

A useful fill report names the instrument, side, total filled quantity, average fill price, fees and resulting position.

Rejected

Rejected means the order did not enter the market.

The rejection may come from the account control layer or the venue. Common reasons include an invalid market, an order above the account limit, insufficient funds, an invalid price or quantity, or disabled trading.

The agent should preserve the reason. It should not quietly reduce the order, switch instruments or claim that the request completed.

If the user asked for $700 and the account cap is $500, the correct result is not a smaller order. It is a rejection.

Cancelled

Cancelled means an open remainder has been removed before it filled.

A cancellation may happen because the user asked for it, the agent changed the plan or the venue expired the order. If part of the order filled first, cancellation applies only to what remained.

The final report therefore needs both facts: how much filled and how much was cancelled.

Tool success is not trading success

This distinction is easy to miss in an AI interface.

A tool call can succeed at the protocol level because the server returned a valid response. That response may contain a rejected order.

The model has to read the business result inside the response. The product also has to return that result clearly enough for the model to understand it.

A green check beside a tool call should not be interpreted as a filled trade unless the returned order state says so.

What an agent should report

After an order request, the agent should be able to answer:

  1. Which exact instrument was used?
  2. What side, size and order type were submitted?
  3. Was the request blocked before the venue?
  4. What order identifier came back?
  5. Is the order rejected, open, partially filled, filled or cancelled?
  6. How much filled and at what average price?
  7. Is any quantity still open?
  8. What changed in the account?

If the state can change after the first response, the system also needs a way to read the order again rather than treating the initial acknowledgement as final.

A simple test

Submit one small limit order away from the current market.

The expected result is not a fill. It should appear as open. Ask the agent to read it back, cancel it and confirm that the remaining quantity is no longer active.

Then submit a small order that can execute and verify the final fill against the account.

This test separates three capabilities that often get blurred together: submitting an order, reading its state and reporting the outcome accurately.

How this applies to agentic trading

Natural language makes the front of the workflow easier. It does not change the venue’s order lifecycle.

The agent can translate an instruction into a tool call. The account and venue still decide whether the request is allowed, accepted and filled.

At Fensory, the useful first step remains an account read. Check the balance, positions and open orders before requesting any action. After an order, ask for the returned state rather than accepting “done” as the answer.

Explore Fensory

Sources


Why order submitted does not mean order filled was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

AI Crypto Trading Bots Are Changing Automated Trading. What Does It Take to Build One?

7 October 2026 at 09:52

Crypto markets move continuously, and manual trading becomes difficult when opportunities appear and disappear within seconds. This is one reason AI-powered trading bots are gaining attention among traders, investment firms, and businesses looking to automate crypto strategies.

But building an AI crypto trading bot is not simply a matter of connecting an AI model to an exchange API. A useful trading system needs reliable market data, strategy logic, risk controls, exchange integrations, execution infrastructure, and continuous monitoring.

So, what does it actually take to build one?

What Is an AI Crypto Trading Bot?

An AI crypto trading bot is an automated software system that analyzes market information and makes trading decisions based on predefined strategies, machine learning models, or a combination of both.

Traditional bots usually follow fixed rules.

For example:

If Bitcoin drops below a certain price, place a buy order.

An AI-powered bot can work with a broader set of signals, including price movements, trading volume, volatility, historical patterns, and other market indicators. Depending on the system, machine learning models can help identify patterns and adjust trading decisions based on changing market conditions.

The important point is that AI should support the trading strategy rather than replace proper risk management.

Why Businesses Are Moving Toward AI-Based Trading?

Automated trading already removes much of the repetitive work involved in monitoring markets. AI adds another layer by helping systems process larger amounts of information and identify patterns that may be difficult to evaluate manually.

Businesses typically consider AI trading bots when they want to:

  • Monitor markets around the clock
  • Execute trades without manual intervention
  • React faster to market movements
  • Automate complex trading strategies
  • Analyze large amounts of market data
  • Reduce emotional decision-making
  • Manage multiple exchanges or trading pairs

However, automation alone does not make a trading bot effective. The architecture behind the system has a major impact on execution speed, reliability, scalability, and risk.

What Does It Take to Build an AI Crypto Trading Bot?

1. A Clearly Defined Trading Strategy

The first step is deciding what the bot is actually supposed to do.

Different businesses may require completely different strategies, such as:

  • Arbitrage
  • Market making
  • Grid trading
  • Scalping
  • Trend following
  • Quantitative trading
  • AI-driven predictive trading
  • Portfolio rebalancing

The strategy determines the type of data the bot needs, how signals are generated, and how orders are executed.

Trying to build the technology before defining the trading logic often results in an expensive system without a clear purpose.

2. Reliable Market Data

AI models are only as useful as the data they receive.

A trading bot may need real-time and historical information such as:

  • Price data
  • Order book depth
  • Trading volume
  • Candlestick data
  • Market volatility
  • Liquidity information
  • Trade history

The system also needs to process this data quickly enough to make decisions without introducing unnecessary delays.

For businesses operating across multiple exchanges, data normalization becomes even more important because each exchange can structure its API and market data differently.

3. AI and Machine Learning Models

This is where an AI trading bot can go beyond simple rule-based automation.

Depending on the use case, developers may integrate machine learning models for pattern recognition, market classification, signal generation, sentiment analysis, or predictive analytics.

For example, an AI model could evaluate several market indicators simultaneously and assign a probability to a potential price movement.

But the model should not be treated as an automatic profit generator.

It needs to be trained, tested, evaluated, and continuously monitored. Poor-quality data or overfitting can produce signals that look strong during testing but fail in live markets.

4. Exchange API Integrations

A trading bot needs direct communication with crypto exchanges to retrieve market data and execute orders.

The integration layer generally handles:

  • Market data retrieval
  • Account information
  • Order placement
  • Order cancellation
  • Order status tracking
  • Balance monitoring
  • Authentication

If the bot is designed for multiple exchanges, the architecture should make it possible to add new integrations without rebuilding the entire trading engine.

5. Risk Management

This is one of the most important parts of the system.

A sophisticated AI model does not eliminate trading risk. The bot still needs controls that determine how much capital can be exposed to a position.

Risk management can include:

  • Stop-loss rules
  • Take-profit conditions
  • Position sizing
  • Maximum daily loss
  • Exposure limits
  • Trade frequency limits
  • Drawdown controls
  • Emergency trade shutdowns

A good trading bot should have the ability to stop trading when predefined risk conditions are reached.

6. Automated Trade Execution

Once a trading signal is generated, the system needs to convert that signal into an actual order.

Execution logic needs to consider factors such as:

  • Market liquidity
  • Slippage
  • Order type
  • Available balance
  • Transaction fees
  • API response time
  • Partial fills

For high-frequency strategies, even small execution delays can affect the outcome of a trade.

This makes the trading engine and execution infrastructure just as important as the AI component.

Backtesting Before Going Live

One of the biggest mistakes in automated trading development is taking a strategy directly from an idea to live trading.

Backtesting allows developers to evaluate how a strategy would have performed using historical market data.

A useful testing process can examine:

  • Historical returns
  • Drawdowns
  • Win and loss ratios
  • Trade frequency
  • Slippage assumptions
  • Transaction costs
  • Market volatility
  • Performance across different market conditions

The goal is not to prove that a strategy will always make money. It is to identify weaknesses before real capital is exposed.

After backtesting, businesses can use paper trading or simulation environments to evaluate the bot under live market conditions without immediately risking funds.

Security Cannot Be an Afterthought

Because a crypto trading bot may have access to exchange accounts and trading permissions, security needs to be built into the architecture from the beginning.

API credentials should be handled securely, unnecessary withdrawal permissions should be avoided, and sensitive information should not be exposed through application logs or frontend interfaces.

Other important considerations include:

  • Encrypted credential storage
  • Secure API authentication
  • Role-based access
  • Server-side validation
  • Activity monitoring
  • Audit logs
  • Infrastructure security
  • Emergency access controls

The more capital a platform is expected to manage, the more important these controls become.

A Scalable Architecture Matters

A basic trading bot may work with one exchange and a few trading pairs. A commercial platform may need to support multiple exchanges, strategies, users, and portfolios simultaneously.

A scalable architecture can separate important components such as:

Market Data → AI/Strategy Engine → Risk Engine → Order Management → Exchange APIs → Monitoring

This separation makes it easier to test individual components, introduce new strategies, add exchanges, and improve the system without disrupting the entire platform.

What Is the Cost of Building an AI Crypto Trading Bot?

There is no single development price because the scope can vary significantly.

A simple bot supporting one exchange and a limited strategy will require considerably less development than a multi-exchange AI trading platform with advanced analytics, portfolio management, risk controls, and an administrative dashboard.

Development cost can depend on:

  • Number of exchanges
  • Number of trading strategies
  • AI model complexity
  • Real-time data requirements
  • User and portfolio management
  • Security requirements
  • Trading dashboard
  • Backtesting infrastructure
  • Mobile or web applications
  • Third-party integrations

For this reason, businesses should define the trading workflow and technical requirements before estimating the actual development budget.

Should You Build an AI Trading Bot From Scratch?

That depends on the business model.

A custom-built solution makes more sense when the platform requires proprietary strategies, specialized AI models, custom risk controls, or integrations that standard trading software cannot provide.

For businesses that want to launch faster, a prebuilt foundation or modular approach can reduce development time while still allowing important components to be customized.

The right approach depends on whether the priority is speed to market, control, scalability, differentiation, or long-term ownership of the technology.

Choosing the Right Development Partner

The development team matters because an AI crypto trading bot combines several technical areas: blockchain, exchange APIs, trading systems, AI, data engineering, cybersecurity, and backend infrastructure.

When evaluating a development partner, look beyond whether they can build a dashboard or connect an exchange API.

Ask whether the team understands:

  • Automated trading architecture
  • Exchange API integration
  • AI and machine learning
  • Trading strategy implementation
  • Risk management
  • Backtesting
  • Real-time data processing
  • Security
  • Scalable backend infrastructure

A development partner should be able to translate the trading concept into a technical architecture that can actually operate under live market conditions.

For businesses planning a custom solution, working with a team experienced in crypto trading bot development can help turn the trading strategy into a complete automated platform rather than just a basic trading script.

Final Thoughts

AI is changing what automated crypto trading systems can do, but the AI model is only one part of the equation.

A successful trading bot needs reliable market data, a well-defined strategy, strong risk controls, secure exchange integrations, efficient execution, testing infrastructure, and continuous monitoring.

The real challenge is not simply adding AI to a trading bot. It is building an entire system where AI, trading logic, execution, security, and risk management work together.

For businesses considering an AI-powered trading platform, defining these components before development can make the difference between a prototype that looks impressive and a trading system designed for real-world operation.


AI Crypto Trading Bots Are Changing Automated Trading. What Does It Take to Build One? was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

❌
❌