Fuzzwah's Agentic Coding Learnings

Five months between posts. In my defence, the whole setup got rebuilt in that time, and I wanted to write about the thing I landed on rather than the thing I was excited about for a fortnight.

The last post ended with a line I was pretty proud of: “One machine. One brain.” That was the split-brain conduit setup, and it genuinely was a good answer to the problem I’d been having. Sounded like a conclusion. Turns out it was more of a stopping point — the kind you hit when you’ve solved the problem in front of you...
Continue reading →

For a while there, my Claude Code setup was doing my head in.

I’ve got two MacBooks — one at my desk, one on the couch — and I was bouncing between them depending on where I felt like working. Fine in theory. In practice, it meant agent sessions were scattered across both machines in various states of completion. Half-finished features on one, a separate chain of thought on the other. Merging the work back together was tedious, and that’s the optimistic framing.

The deeper problem was the harness state. Claude Code writes memory files to the local filesystem, and...
Continue reading →

Here’s a thing I keep doing that I didn’t plan to turn into a habit: every now and then, instead of giving an AI agent a task, I ask it to explain my project back to me.

That sounds like a waste of a perfectly good agent session. It’s not. It’s become one of the most useful things I do.

The Setup

I run multiple AI coding agents in parallel across different repos. Each one is bootstrapped with a CLAUDE.md file — a markdown document that gets loaded automatically at the start of every conversation. It’s the agent’s institutional...
Continue reading →

There’s a point in any project where the number of terminal tabs starts to feel like a personal failing. I’d been running Claude Code sessions across six different repos — switching between terminals, trying to remember which agent was doing what, occasionally discovering that two agents had different ideas about the database schema. It worked, technically. The same way juggling chainsaws works until it doesn’t.

I needed a system, not just more terminal tabs.

The Mess

The platform I’m building has six Git repositories:

• A Discord bot (Python, py-cord) that handles user interactions
• A Django web app (DRF,...
  Continue reading →

The intro post from last week covered what StackTUI does and how to try it. This one’s for the people who came back asking about the how — the Textual architecture, the threading model, what a real config file looks like, and how you actually wire it up to your own project. There’s also a proper look at the OpenSpec + Claude Code workflow that was used to build it, which I keep mentioning in passing and should probably explain properly at least once.

What You’re Looking At

When StackTUI opens you get a four-zone layout. At the...
Continue reading →