Agents and Tools

The Idea

An agent is a model in a loop with tools. You describe each tool — name, purpose, parameters — and the model answers with a structured call instead of prose. Your code runs it, returns the result, and the loop repeats.

Real-World Example

A field surveyor with a radio. When the map disagrees with the ground, they call the office for one measurement, wait for the number to come back, and only then carry on drawing.

The Tradeoff

Tools break the model out of its training data and its shaky arithmetic. Every extra step also multiplies latency, cost, and the chance of one wrong turn compounding into five. Loops need hard caps and tools need permission checks, because fetched text can try to steer the agent.

def get_invoice(invoice_id):
  return {"id": invoice_id, "total": 90}

tools = {"get_invoice": get_invoice}
call = {"name": "get_invoice", "arguments": {"invoice_id": "INV-42"}}

# hand the model's arguments to your function
result = tools[call["name"]](___)
print(result)     # want {'id': 'INV-42', 'total': 90}