The agent loop

An agent is a model in a loop: it reads context, decides to call a tool or answer, receives tool results, and repeats until done.

Stable: this rarely changes. Reviewed 17 Oct 2026.

Key ideas

Core loop: call model, if it returns tool calls execute them, append results, call again; stop on a final answer or a limit.
Always set guards: max turns, max tokens, wall-clock timeout, cost budget.
Run independent tool calls in parallel when the model emits several.
Return tool errors to the model as results so it can recover; do not crash the loop.
Log every step (input, tool call, result, tokens) for debugging and evals.

In code

python

messages = [{'role': 'user', 'content': task}]
for turn in range(MAX_TURNS):
    resp = client.messages.create(model=MODEL, tools=TOOLS,
                                  messages=messages, max_tokens=4096)
    messages.append({'role': 'assistant', 'content': resp.content})
    if resp.stop_reason != 'tool_use':
        break
    results = [run_tool(b) for b in resp.content if b.type == 'tool_use']
    messages.append({'role': 'user', 'content': results})

Common mistakes

Using an agent where a fixed workflow would do. Agents cost more and are harder to test.

The agent loop

Key ideas

In code

Common mistakes

Go deeper