2026-03-26
The slot machine myth: ChatGPT compared AI output quality to a slot machine. I genuinely don't get it. Output is predictable when you obsess about what goes in – your CLAUDE.md (persistent context), your documents, your session discipline. The bottleneck shifted from capability to context a while ago. Most people are still blaming the model. Tweet
Cowork vs Claude Code: The takes keep coming – terminal vs app, which is better. The point is whether your instructions and workflows compound across sessions. That's what separates an assistant from a system. Blog post
Output styles vs CLAUDE.md: CLAUDE.md appends to the default (coding-oriented) instructions. Output styles replace them. That's the difference between fighting defaults and changing what the tool is. Tweet
"I'm not a terminal person": I keep having this conversation with founders about Claude Code. Four simple commands and you're able to chat just like Cowork. The moment it clicks is the second session – Claude already knows their company. I wrote the walkthrough so I can stop repeating myself. Blog post
Content entropy is becoming my favourite underrated problem. AI defaults to adding, not subtracting. Every edit quietly expands what you didn't touch. Bloated docs become context for the next session – and then the next session's output is slightly worse and you don't know why. Tweet
Your sessions are too long. One task, commit, exit, new session. It sounds painful. It isn't. New sessions are 'free' when your CLAUDE.md is solid. Tweet (sessions) · Tweet (compaction)
Fresh starts over long sessions. Anthropic now recommends starting new conversations instead of compressing old ones. I've been running this way for months. It's nice to see it in the official docs. Docs
The 1M context window is headroom, not a target. It's how much context the AI can hold in one conversation. More isn't better – bigger context means more noise, less focus. The sweet spot is around 15-30%, then start fresh.
The AI Vampire: Steve Yegge asks whether heavy AI usage costs more energy than it saves. He articulates what a lot of us think privately but don't post about. I don't agree with his conclusion, but I liked the question. Essay
Don't miss what's next. Subscribe to Build Notes: