The method
The Architect's Playbook
The best results do not come from clever tricks. They come from treating AI as a smart but scattered collaborator that needs structure, clear goals, and honest feedback. The goal is never to get lucky with one output — it is a repeatable system that produces excellent thinking on demand.

The seven-step framework
Applied at the end of almost every important prompt.
- 01Restate the goal in your own words.
- 02Identify any missing information you still need.
- 03Explicitly list every assumption you are making.
- 04Break the task into the smallest logical sub-steps.
- 05Walk through the reasoning step by step.
- 06Deliver the final answer.
- 07Self-critique: name the weakest part and revise it.
When the framework alone is not enough
Tree of Thoughts
Explore multiple reasoning paths, critique each one, then choose or combine the strongest.
ReAct
Interleave reasoning with action. Think, act, observe, continue — instead of guessing in one pass.
Self-consistency
Generate several independent answers, then synthesize the one the evidence keeps landing on.
Meta-prompting
Have the model critique and rewrite its own prompt before it ever answers the question.
Multi-agent collaboration
Researcher · Critic · Synthesizer
Gather, challenge, integrate. The right shape for strategy and content work.
Planner · Executor · Verifier
Break down, execute, validate. The right shape for projects and code.
Debate
Two opposing agents argue the case, a judge rules. Use it when the stakes are high.
Manager · Specialists
A manager delegates to domain experts and integrates the results. Use it when the work spans domains.
Which approach, and when
Want this installed in your team?
This is the same methodology behind every product on this site, and it is what I teach.
Start a conversation