The same loop, pointed at the markets
Nothing in this unit is a new idea. Take the perceive → decide → act loop you have already met and aim it at a trading screen, and you get an agent using market software the way a professional uses a terminal:
- Perceive — read a chart, an option chain, or the order book as data.
- Decide — form a view of the situation.
- Act — place or manage an order through a tool, then look again and adjust.
The examples here use generic instruments — stand-ins that stand for any real one — because the goal is to teach the shape of the thing, not a specific market or strategy.
The hard part is perception, not the order
It is tempting to think the order is the interesting part. It is the opposite. Placing an order is a single, well-defined action — the easy end. The step most systems get wrong, or skip, is perception.
Picture a trader at a terminal: they see the whole screen — price, depth, volatility, the shape of the moment — and read it at a glance. Now picture how many systems hand an agent its view of the market: a thin slice of blind data through an API, a few numbers with no gestalt.
An agent given blind data trades blind. It can still act — the order tool works fine — but it is acting on something it cannot really see. That is the decisive gap. Getting perception right, so the agent sees what a professional sees, is the hard and important part; the order is the trivial one.
See it in kestrel
Kestrel's whole premise is the perception step done properly: it renders a session as something an agent can actually read, not a blind API slice. Watch one on a generic instrument, no signup, no card:
npx kestrel.markets sim fomc-rate-decision-whipsawIt prints a certified proof URL you can recompute yourself. Drop the kestrel.markets MCP server into your client and the screen is there again next session — no account in between.