Advanced70 minBuilding AI Agents

Build a ReAct Agent Loop

Implement the ReAct (Reason + Act) loop: the model reasons, picks a tool, observes, and loops until done.

Scenario

Build the agent loop from scratch — no framework. Understand exactly how agents work by implementing one.

Objective

Implement run_agent(task, tools, max_iter) that loops reasoning + tool calling until final answer or max_iter.

Starter code

Implement the TODOs to complete the lab.

agent.pypython

def run_agent(task: str, tools: list[Tool], max_iter: int = 10) -> str:
    """Run the ReAct loop.
    - Call LLM with task + tool definitions
    - If model returns tool_calls, execute them
    - Feed observations back to model
    - Loop until final answer or max_iter
    - Raise on max_iter exceeded
    """
    messages = [{"role": "user", "content": task}]
    # TODO: implement the loop
    pass

Solution hints

1Pass tools as function definitions
2Check response.choices[0].message.tool_calls
3Execute each tool, append tool result as 'tool' role message
4Stop when no tool_calls and finish_reason='stop'
5Enforce max_iter strictly

Validation steps

Your implementation should pass these checks.

Agent calls the right tool for the task
Agent loops until it has the answer
Agent stops at max_iter with an error
Tool errors are caught and fed as observations

Run validation

This is a simulated validation environment. In production, this would run your code against the validation steps.