"AI trading agent" now covers very different products. Some suggest trades, some decide on their own, some execute rules you wrote. Connected to your exchange, the difference between them is the difference between a tool and a gamble.
What an AI trading agent is
An AI trading agent is software that uses an AI model to read market information and act on your exchange or broker account: open and close positions, set stops, manage orders. It connects through the exchange's API, sometimes through an MCP connector, using a key you create.
What changes from one agent to another is a single question: who decides the trade?
The 3 types
1. The assistant that suggests
A chatbot that analyses a chart or the news and proposes an idea. You decide and you click. It is useful for research, but the result depends on your discipline in the moment, and its suggestions change with how you phrase the question.
2. The autonomous agent that decides
The agent has its own logic, often opaque, and trades on its own. You are trusting a strategy you cannot read, test or explain. When it loses, you cannot tell whether the market changed or the agent improvised. This is the type that tends to come with return promises.
3. The agent that executes your rules
You write the strategy, it is turned into precise rules and backtested, then the agent checks the rules and executes exactly what they say. No rule met, no trade. The AI does the work that is hard for a human (watching, being consistent, logging), not the thinking that should stay yours.
| Question | Suggests | Decides alone | Executes your rules |
|---|---|---|---|
| Who picks the trade? | You | The agent | Your rules |
| Can you test it first? | Hardly | No | Yes |
| Can you explain each trade? | Sometimes | Rarely | Always |
| Consistent over time? | Depends on you | Unknown | Yes |
The risks, whatever the type
- Your funds. An agent should never be able to withdraw. Trading permission only on the API key.
- Your risk. Position size, leverage and maximum loss must be capped by code, outside the AI.
- What it reads. An agent that reads the web or messages can be manipulated by hidden instructions in that content (prompt injection).
- Promises. Nobody can guarantee trading returns. A promise of returns is the clearest warning sign there is.
7 questions before you connect an agent to your account
- Who decides each trade: me, my rules, or the agent?
- Can I read the exact rules it follows?
- Were they backtested with fees, funding and slippage, on data not used to design them?
- Does the API key exclude withdrawals?
- Are risk limits enforced outside the AI?
- Is every order logged with the reason that triggered it?
- Does anyone promise me returns? If yes, stop there.
Anyone offering to manage your money for you should also be authorised by the financial regulator in your country. Check before you connect anything.
Where Sunia fits
Sunia is the third type. You describe your strategy in your own words, the agent turns it into precise rules and asks when something is unclear, the rules are backtested, then it trades them on your exchange or broker through an MCP connector, with every trade logged. It never invents a strategy of its own, never holds your funds, and your keys cannot withdraw. More on the connection itself in our guide to MCP trading.