Lunch Hunch

I’ve recently listened to a couple of highly motivating podcasts — DHH on the Pragmatic Programmer and Simon Willison on Lenny’s. Both challenged me to be more ambitious in what I’m building and to embrace experimenting more.

I’ve been a bit worried about my coding skills atrophying since adopting an agent-first approach. My days are no longer spent poring over the shape of methods, classes and tests — instead they’re spent thinking about what to build and the high-level approach to doing so. I’ve noticed my ability to write code from scratch has quietly slipped — it’s just not what I’m doing most of the day anymore.

The constraint

I generally have quite a few ideas, as do most programmers, but finding the time — that’s a different matter. I have a very active two-year-old who takes up most of my time and energy outside of work. He is my side project, if you will.

The thing I realised

Agentic AI — Claude Code, Codex, whatever you’re using — completely changes this.

I’ve started having what I call a “lunch hunch”: there’s something I want to explore, so I have a quick chat with an agent to give it a rough idea of what I’m thinking — it’s amazing how unformed the ideas can be — then I come back from lunch and see what’s been produced. I can kick it off again, get on with more important things, and come back to it later. The ease of using Git worktrees in Codex or Claude Code has made this pretty trivial.

Hunches this month

None of these would exist without AI coding tools. Pre this AI era, I’d never have even attempted them.

• A Slackbot that reacts to messages and proposes updates to our database at work
• A feature that watches for mentions of entities in the media and notifies us
• Explored Next.js and FastAPI by building my perennial side project — a personal app that stores recipes in a consistent format from around the web
• Explored Astro by building this revamped personal site

What I’ve learned

A lot of my ideas are very unformed, so the output will be far from perfect — and that’s fine. It’s about exploring. I’m not shipping this to production tomorrow. I just have itches I want to scratch and I’m curious what’s possible.

A lot of people think the cost of these tools will go up considerably and we’re in a golden age right now. I’m keen to make the most of it while tokens are relatively cheap and I have the ideas to pursue.

I’ve started messing about with Claude Code on mobile and can see how the “lunch hunch” might become the “all day hunch”.