Plan a Project
You have learned decomposition, subgoals, milestones, sequencing, adaptive replanning, backtracking, effort estimation, and planning agent architecture. Now it is time to use all of those tools at once — to build a complete, realistic multi-step plan the way a skilled AI agent would. This lesson is a workshop. Minimal explanation, maximum doing.
Planning is a skill, not just a concept. The first plan you make will be rough. The tenth will be dramatically better. Real-world AI planning researchers iterate constantly — they build plans, run them, observe where they break, and rebuild. You are doing the same thing here.
The Planning Challenge
Your challenge is to build a complete plan for one of the following goals. Choose the one that interests you most, or propose your own goal of similar complexity. Option A: Launch a student-run online store selling handmade bookmarks, with at least 10 products listed and the first order fulfilled within 30 days. Option B: Produce a 10-minute documentary video about a person, place, or topic in your community, from concept to published online. Option C: Organize and run a school debate tournament with at least three rounds and six teams, from first announcement to trophy ceremony. These goals are deliberately large and realistic — each involves dozens of tasks, real dependencies, and meaningful uncertainty. You are not just listing steps; you are building a plan that could actually guide a real project.
Build Your Full Project Plan
- PHASE 1 — DECOMPOSE (10 minutes)
- Decompose your chosen goal into 4 to 6 major subgoals. Write each as a clear, noun-phrase label: Research, Design, Produce, Test, Launch. Then decompose each subgoal into 3 to 5 specific leaf tasks — tasks concrete enough to start immediately with no further planning.
- PHASE 2 — SEQUENCE AND MAP DEPENDENCIES (10 minutes)
- For each leaf task, identify any tasks that must be complete before it can start. Draw a dependency diagram (boxes and arrows). Then find which tasks can run in parallel — mark them with the same color or a P symbol.
- PHASE 3 — SET MILESTONES (5 minutes)
- Choose 3 to 4 milestones from your decomposition — the most significant checkpoints. Write each as a concrete, verifiable statement: All products photographed and descriptions written, not Working on product listings.
- PHASE 4 — ESTIMATE EFFORT (5 minutes)
- For each leaf task, write a time estimate in hours or days. Use the planning fallacy defense: after you write each estimate, multiply it by 1.5 to account for hidden complexity and realistic interruptions.
- PHASE 5 — ADD CONTINGENCIES (5 minutes)
- Identify the 2 most likely failure points in your plan. For each, write an if-then contingency: If [failure], then [adjusted plan].
- PHASE 6 — REVIEW AND REFINE (5 minutes)
- Read your plan from start to finish. Identify any step where you are not sure exactly what to do — that step needs further decomposition. Identify any two tasks whose estimated total time exceeds the 30-day window — these need to be parallelized or descoped. Make the fixes.
- DELIVERABLE: Your finished plan should include: the goal statement, a decomposition tree with at least 15 leaf tasks, a dependency diagram, 3 to 4 milestones with verifiable statements, effort estimates for each task, and 2 contingency branches. Present it as clearly as you can — a real agent, or a real project manager, should be able to follow it.
What Makes a Plan Excellent?
After you have finished your plan, evaluate it against these four criteria. Completeness: are there any obvious gaps — things that clearly must happen but are not listed? Verifiability: can you check each milestone with a clear yes or no? Realism: are the effort estimates plausible, not optimistic fantasies? Flexibility: do your contingencies give the plan a genuine path forward if something breaks? A plan that scores well on all four criteria is what a well-designed planning agent strives to produce before executing any action. The discipline of building this plan — the careful thinking before the doing — is exactly what separates effective agents from impulsive ones.
When evaluating a project plan, what does verifiability mean?
A student finishes a project plan and realizes two of the tasks, each estimated at 15 hours, have no dependency on each other. The total project window is 30 hours. What should they do?
The ability to build a complete, realistic, flexible plan before acting is one of the rarest and most valuable skills in any field — engineering, medicine, business, research, film production. It is also one of the core capabilities that makes AI agents genuinely useful rather than just impressive. Every lesson in this module has been building toward exactly this skill.