Skip to content
Hands-on AI Playbook

Planning

What It Is

Section titled “What It Is”

Planning is a pattern where an agent decomposes a complex goal into a sequence of smaller, manageable steps — then executes them in order. Instead of attempting to solve a problem in a single response, the agent first creates a plan, then follows it step by step, adapting as needed based on intermediate results.

This is what separates an agent from a simple chatbot. A chatbot responds to a single prompt. An agent with planning capability can take a high-level objective (“process this customer’s exchange”) and autonomously determine the sequence of actions required to achieve it.

Why It Matters

Section titled “Why It Matters”

Real-world tasks are rarely single-step. Processing an exchange requires verifying the order, checking return eligibility, confirming inventory, processing payment, and sending confirmation. A planning agent handles this entire workflow without requiring the user to specify each step.

Andrew Ng has noted that planning is the least mature of the four core agentic patterns — it works well for well-defined workflows but remains challenging for open-ended, ambiguous goals. This makes it both the most powerful pattern (when it works) and the one that most benefits from guardrails and human oversight.

How It Works

Section titled “How It Works”
┌──────────┐     ┌──────────────┐     ┌──────────────┐
│  Goal     │────▸│  Plan         │────▸│  Execute      │
│  (input)  │     │  (decompose)  │     │  (step by    │
│           │     │               │     │   step)       │
└──────────┘     └──────────────┘     └──────┬───────┘
                                             │
                                      Replan if needed
  1. Receive goal — The agent receives a high-level objective.
  2. Decompose — The agent breaks the goal into an ordered sequence of sub-tasks.
  3. Execute — The agent works through each sub-task, using tool calls and reflection as needed.
  4. Monitor — After each step, the agent checks whether the result changes the remaining plan.
  5. Replan — If a step fails or produces unexpected results, the agent revises the remaining plan rather than blindly continuing.

Advanced planning approaches include:

  • Chain-of-thought planning — The agent reasons through the plan in natural language before executing.
  • Tree-of-thought — The agent explores multiple possible plans and selects the best one (Yao et al. 2023).
  • Hierarchical planning — High-level plans are broken into sub-plans, each with their own steps.

Example

Section titled “Example”

Customer exchange scenario

Section titled “Customer exchange scenario”

Goal: “Process customer Jane Smith’s exchange — return Blue Widget, ship Red Widget.”

Agent’s plan:

  1. Look up Jane Smith’s order history → find order #ORD-5678
  2. Verify return eligibility → check if Blue Widget is within the 30-day return window
  3. Check Red Widget inventory → confirm availability in the nearest warehouse
  4. Calculate price difference → Blue Widget was $29.99, Red Widget is $34.99, difference is $5.00
  5. Process return for Blue Widget → generate prepaid return label
  6. Charge $5.00 price difference → process payment
  7. Place order for Red Widget → create new shipment
  8. Send confirmation email → include return label, new order details, and timeline

Replanning example: At step 3, the Red Widget is out of stock. The agent replans: offer the customer the Green Widget as an alternative, or place the Red Widget on backorder with an estimated date.

Research task

Section titled “Research task”

Goal: “Write a competitive analysis of our top 3 competitors.”

Agent’s plan:

  1. Identify the top 3 competitors from the company’s CRM data
  2. For each competitor, gather recent news, product launches, and pricing
  3. Analyze strengths and weaknesses relative to our product
  4. Draft a comparison table
  5. Write executive summary with recommendations
  6. Review and refine the full document

When to Use It

Section titled “When to Use It”
  • Multi-step workflows with dependencies between steps (step 3 depends on step 2’s output)
  • Tasks where the sequence of actions matters
  • Goals that are too complex to accomplish in a single tool call
  • Workflows where failure at one step should change the approach for subsequent steps
Section titled “Related Patterns”

Further Reading

Section titled “Further Reading”