Planning

Without planning, agents react step-by-step with no global view — they backtrack, repeat work, and miss dependencies. Planning front-loads reasoning: 'To book a trip, I need flights first, then hotel, then car rental.' This reduces wasted tool calls and improves success rate on multi-step tasks.

Planning is the difference between a tourist wandering a city randomly versus following a curated itinerary — both might see sights, but the planner hits every must-see efficiently.

Scenario

Goal: 'Prepare quarterly board report.' Plan: (1) Query sales DB for Q3 metrics, (2) Generate charts from data, (3) Draft executive summary, (4) Format as PDF, (5) Email to board@company.com. Executor runs each step sequentially, re-plans if Q3 data is incomplete.

Solution

In Agent Foundations, apply Planning to this scenario: Goal: 'Prepare quarterly board report. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Planning (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Planning happens in the code.

1PLANNER_PROMPT = """Create a step-by-step plan to accomplish this goal.  # key line for Planning2Return JSON: {"steps": [{"id": 1, "description": "...", "tool": "tool_name", "depends_on": []}]}  # key line for Planning3
4Goal: {goal}5Available tools: {tools}"""6
7def create_plan(goal: str, tools: list) -> list[dict]:  # define a reusable function8    response = client.chat.completions.create(  # call the API9        model="gpt-4o",10        messages=[{"role": "user", "content": PLANNER_PROMPT.format(goal=goal, tools=tools)}],  # key line for Planning11        response_format={"type": "json_object"},12        temperature=0,13    )14    return json.loads(response.choices[0].message.content)["steps"]  # return the result15
16def execute_plan(plan: list[dict], tools: dict) -> str:  # define a reusable function17    results = {}18    for step in plan:  # key line for Planning19        deps = [results[d] for d in step.get("depends_on", [])]20        result = tools[step["tool"]](step["description"], *deps)21        results[step["id"]] = result22    return results  # return the result

• Rigid plan with no re-planning on failure
• Over-planning simple tasks — adds latency for no benefit
• Not validating plan feasibility before execution
• Plans with circular dependencies
• Planner and executor sharing no context about failures

Quick recap — the most important points from this module.