AI trading covers everything from a model that scores a stock to an agent that reads the market and trades on its own. Here is what each kind actually does, what the evidence says about results, and how to start without losing money to a mistake you could have tested first.
November 16, 2026
10
min to read
The short answer
AI trading means using artificial intelligence, from simple scripts to language models and full AI agents, to read markets, test ideas and place orders. It covers machine-learning signals, chatbots you ask for research, AI agents that act within limits you set, and rule-based bots with AI on the label. None of them can reliably predict prices.
AI trading means using artificial intelligence, in some form, to help with part of buying and selling: finding an idea, reading a chart, testing a rule, watching a position, or sending the order itself. The phrase stretches over a lot of different software. A model that scores a stock from years of data, a chatbot you ask a question, and a program that reads the market and places an order on its own are all sold as “AI trading” today, and they work in very different ways.
What ties them together is the kind of intelligence doing the work. A machine-learning model learns a pattern from historical data and turns it into a number: a score, a probability, a shape it noticed on a chart. A large language model, the kind of AI behind Claude and ChatGPT, reads and writes instead: it can sum up a filing, turn your sentence into a rule, or answer a question about a chart you show it. Put either kind of model inside a program that keeps watching the market on its own and can act on what it finds, and you get an AI agent. And plenty of software marketed as an “AI trading bot” uses none of the above: a fixed rule places the trade, and the AI, where there is any, only suggests a setting.
None of this is new in spirit. Traders have used statistics to time the market, and computers to place orders from fixed rules, for decades. What changed is what now counts as “using AI”: cheaper machine learning, and language models that can read a plain-English request and act on it.
Start by asking what is actually making the call: a model trained on data, a language model you are talking to, an agent working on its own, or a fixed rule someone wrote by hand.
| Kind | What decides | What you get | |
|---|---|---|---|
| Machine-learning models | A model trained on past prices, fundamentals or news | A score, a signal or a pattern flagged on a chart, for you or a rule to act on | AI stock trading |
| Language-model assistants | You, after it reads, summarizes or drafts for you | An answer, a chart read or a drafted order you place yourself | Claude and ChatGPT |
| AI agents | The agent, inside the limits you set | Research, alerts and orders, anywhere from informing you to acting on its own | what an AI trading agent is |
| AI-labeled bots | A fixed rule; the AI, if any, only suggests a setting | Predictable, testable automation, such as a grid or a scheduled buy | the best AI trading bots |
The four overlap in practice. A good agent often turns your idea into a fixed rule before it runs it, which borrows the bot’s predictability. A language-model assistant with the right tools connected becomes part of an agent. And a “bot” with a machine-learning signal bolted on is really the first and fourth kinds stapled together. Agentic trading is the name for the broader move, across brokers and exchanges in 2026, toward letting an agent act for you rather than just answer a question.
Strip away the branding and almost every AI trading tool runs the same five-stage pipeline. Weak at any one stage, and the whole thing is weak.
Prices and candles, news, your account or your open positions, pulled through an API (the interface programs use to talk to an exchange or a broker) or a web search. Nothing downstream can be more current than this feed.
A machine-learning model scores what it sees against patterns in its training data. A language model reads the same data in plain language and reasons about what matters, the way you would read a chart yourself, only faster and without getting tired.
Either a single number, such as a score or a “buy”, or a full plan you can check line by line: the market, the entry, the exit, the size. A plan you can read is a plan you can test before it costs you anything.
You place the order by hand after reading the signal, an assistant drafts it for your approval, or an agent sends it to a broker or an exchange on its own, inside limits set in advance.
A fill, a profit or a loss, and, in the better tools, a record of why the trade happened. Review is how a rule earns the right to keep running, or gets switched off before it does more damage.
Turning a vague idea into the exact rules a computer can run at step three, and the path from there to something live, is its own skill. Algorithmic trading for beginners walks through it with a worked example, a real year of results and a one-line bug that flatters a backtest. When an AI assistant runs that test for you, our guide to AI backtesting covers when it actually computes the numbers and when it only describes them.
Not the way the marketing suggests. Machine-learning models do find real patterns in market data, but the predictable part tends to be small, and it tends to shrink once enough traders lean on the same signal. Language models read a mountain of text fast, which is not the same skill as knowing what a price will do next. Our guide to whether AI can predict the stock market goes through the peer-reviewed research in detail, with a test on real market data you can repeat yourself.
That does not make AI trading worthless; it changes what it is good for. Our guide to AI stock trading breaks the job down task by task, and the same pattern shows up everywhere: AI is strongest at reading, sorting and testing, where a mistake costs you time, and weakest at the one thing most people actually want from it, which is calling the market’s next move. The US derivatives regulator put the limit plainly in a 2024 customer advisory on AI-labeled trading bots: the technology “can’t predict the future or sudden market changes.” A platform that promises otherwise, with a fixed return or a win rate near 100%, is the clearest sign to walk away; is AI trading legit? covers the rest of the warning signs. If you want to try one, here is what an AI trading app does.
Three risks show up whichever kind of AI trading you use.
Start where a mistake costs time, not money, and bring in real money last.
Screening, a chart read, a backtest, or a scheduled buy. “Trade for me” is not a job you can check; “tell me if gold’s RSI drops under 30” is.
The market, the entry, the exit, the size, and what ends it. If a computer needs to guess what you meant, you have not finished writing it.
Backtest it on past data with real costs included, then run it on paper, which means simulated money at live prices, for a few weeks.
The smallest size the venue allows, for long enough to see the backtest and the live result agree, or to find out why they do not.
If what you actually want is to invest for years rather than trade for days or weeks, the tools and the questions differ too; our guide to AI investing covers robo-advisors, AI stock scores and what changes over a longer horizon. At the other end, our guide to AI day trading looks at trading within a single day and what the research says about it. And paper trading crypto has a 30-day plan for testing on live prices with no money at risk.
We build Ethy, so weigh this section knowing that. Ethy is an AI trading agent you ask before you trade, and it covers more than one of the kinds above in a single chat.
Your balance lives in a wallet that belongs to you, and you can export its key at any time. Ethy is software for research and analysis; it does not give investment advice.
They overlap but are not the same. Algorithmic trading means a program follows exact rules a person wrote and places the orders. AI trading is broader: it includes those rule-based bots, but also machine-learning models that learn a pattern from data, language-model assistants you ask in plain English, and AI agents that combine reasoning with the ability to act. A rule bot is one kind of AI trading, not the whole category.
Not reliably. Machine-learning models find real patterns in returns, but peer-reviewed research finds the predictable part is small and tends to shrink once many traders use the same signal. Language models are good at reading news fast, not at knowing what a price does next. Our guide to whether AI can predict the stock market goes through the studies in detail.
It is safer when you start narrow: one job (a screen, a chart read, a backtest), one written rule, and real trading costs counted before any money is at risk. It gets risky when a model acts on bad information with full confidence, or when a platform promises a fixed return, which is the oldest scam pattern wearing a new label.
A bot usually runs one fixed rule, such as a grid of buy and sell orders or a weekly purchase, exactly as coded; the “AI” in many of them only suggests a setting. An agent reasons from a goal in plain words, reads live data through tools, and can inform you, propose an order, or place one within limits you set. Many agents end up writing a fixed rule and running it, which blends the two.
Ethy is an AI trading agent: it reads charts on four timeframes, finds out why a market moved, turns a plain-English idea into a rule it backtests on real history, and can place or run that rule once you are ready, with stop losses and loss limits you set. The free plan comes with 500 credits and no card. It is software for research and analysis and does not give investment advice.
Trading involves risk, including the loss of your money. Ethy AI is a research and trading tool and does not give investment advice or recommendations. Backtests are simulations on historical data and do not predict future results.
Deploy your AI trading agent in one click.
