Chapter 18: Your Journey
The compass from chapter 1 in one line: pick one tool, feed it real context, review like an architect, keep your history honest. Then take something real, build it, and tell one person.
Explore all posts in the AgentSpek category.
18 posts
The compass from chapter 1 in one line: pick one tool, feed it real context, review like an architect, keep your history honest. Then take something real, build it, and tell one person.
For six months, I told almost no one about how I was working. The isolation was real, and so was the way out: finding the people figuring out AI-assisted development alongside you.
Somewhere in the daily back and forth with AI you stop watching the change and notice you have become it. The chapter calls that metanoia, and it is about who the developer turns into.
You get used to the walls. You stop seeing them as walls and start seeing them as reality. Then AI removes them, and everything you once called impossible becomes simply another problem to solve.
The moment arrives quietly: one day you are writing code, the next you are directing systems that build faster than you can follow. What you are holding stopped being a tool and became power.
I stopped seeing AI subscriptions as paying for tools and started seeing them as paying for time: systems I would not have attempted at all now get built in my spare time.
Working with AI doesn't just teach you new things. It reveals how much of what you 'know' is shallow, contextual, or simply wrong. And it happens at a pace that's psychologically disorienting.
When you have unlimited patience from your AI teammate, you grow more patient with your human teammates. When you can iterate rapidly on ideas with AI assistance, you become less precious about any particular approach with humans.
Sonnet stalled on the math in a Python pipeline. GPT-5 on my phone solved it, then wrote code that did not fit. I was not choosing between AIs. I was assembling a team. A late-2025 cast.
The AI-generated code sailed through review and passed every test, then failed in production when two processes collided. Holding a quality bar when code arrives faster than you can read it.
I built a content system for my Astro blog in an afternoon, and the hours felt dense rather than fast. Why the loop with AI is a spiral, and why review is now where the time goes.
I gave Sonnet 4 my slow blog build overnight and woke to a better fix than mine. Another night it turned my S3 and CloudFront setup into a multi-region design costing hundreds a month.
How I work in agent mode with Claude Code: a CLAUDE.md file as the specification, whole subsystems handed over, and AWS infrastructure built by describing what should exist, not typing it.
I asked Claude to help me understand my own blog's deployment pipeline. What came back was not better code but better questions, and that is what conversational mode is for.
I don't think in fragments. I don't think in lines. I think in systems. In architectures. In complete solutions to complete problems. So I stopped fighting autocomplete and found what actually works: agent mode.
A commit co-authored by Claude raises new questions for Git: what one commit should hold when a day of work takes twenty-five minutes, and how to record who wrote what.
The subscription is not the real cost of AI tools; the time it takes to get good with them is. How to measure the return when the gain is a team that thinks better, not a faster build.
The research spiral got me before the code did: a tab for every AI coding tool, each promising everything. This chapter maps autocomplete, chat, agent mode and autonomous systems, and where to start.