You've watched the demo. The AI built in four minutes what takes you a day. You tried it yourself. The AI confidently produced something that broke in production, suggested patterns you explicitly banned, and helpfully "cleaned up" working code into elegant wreckage.
Now what?
It AI'nt That Hard is the practitioner's account of figuring it out. Not the marketing version. The real one -- a year of building production software with AI agents, documented through specific failures, kitchen-table arguments about token budgets, and the slow reckoning with a job that changed shape while the author was still learning to hold it.
Seveeteen chapters. One arc. From a midnight panic to a working methodology:
- Context cascading that stops the AI re-reading your codebase every session (Chapter 4)
- Workflow protocols that grew from 15 lines to 262, each line earned through disaster (Chapter 5)
- Workspace isolation born from forty-seven merge conflicts on a Monday morning (Chapter 6)
- Contract-first development after three agents built three incompatible implementations of the same feature (Chapter 11)
- Automated PR review that catches what humans miss: absence (Chapter 12)
- Guardrails written after an agent nearly deleted production config as a helpful cleanup (Chapter 15)
- The uncomfortable questions about craft, identity, and what the job becomes when machines do the building (Chapter 17)
This is the journey from self-doubt to a new kind of confidence: harder-won, more honest, and built on a year of specific failures.
If you've been reading The Accidental AI Orchestrator blog series on Medium, the book is the full story behind the posts -- with the domestic texture, the marital dynamics, and the identity questions that a public blog series couldn't contain.
The failures are real. The solutions are real. The questions don't have tidy answers.
