FREE PLAYBOOK · 30 PAGES
Work with AI.
Still think for yourself.
A small field guide for using Claude Code or Codex on real work—without blindly accepting the output.
REQUEST
“Add a safe export button”
CONTEXT PACKET
AGENT
plan → diff
HUMAN REVIEW
Built from real agent sessions while shipping software for an Indian startup with 100,000+ paying customers.
WHAT CHANGES AFTER READING IT
01
Start clean
Know what to give an agent before it writes a line.
02
Stay in control
Split work between agents without losing the bigger picture.
03
Ship with proof
Use a small review loop so fast code does not become fragile code.
THE CORE IDEA
The agent is a teammate, not a slot machine.
Give it a small job, useful context, and a way to prove it worked. The playbook shows the exact loop.
1 / BRIEF
What should change?
Goal, limits, and a clear finish line.
2 / AGENT
Do the small job.
Use the right files and context, not guesses.
3 / REVIEW
Check. Test. Ship.
You decide what is safe to merge.
INSIDE
Only the useful parts.
Context packets — what to give an agent before asking it to act.
Subagents — when separate research, building, and review helps.
Graph thinking — how to map files, rules, and dependencies simply.
Review checklists — how to catch mistakes before they become production problems.
YOUR NEXT FEATURE IS ENOUGH