Don't ask "will AI ruin my job?" Ask "how can I use AI to stay ahead?"

Please @AnthropicAI, @claudeai. Why do we have to keep doing this? Please support AGENTS.md

Mass surveillance is fundamentally incompatible with democracy

While I love coding with GPT and Opus, I run the same code reviews with both on all PRs. Have done so for months. Opus catches a LOT more issues than GPT. Large dataset now and the difference is really clear. GPT is better for many things but NOT code review.

New Twitter algo just dropped. It's now Grok based. Many of the rules have changed github.com/xai-org/x-algorit… As this is very relevant to nearly all of us, I'll break it down a bit: 1. Optimise for a clear audience. Don't write for everyone. Be targeted. 2. Make the first line in your tweet stop people scrolling and 'spend time' on your post. 3. Add structure to create dwell time. 4. Replies are really important. Drive replies, not just likes. 5. Media really helps. Add images when you can. Especially ones people stare at for ages. 6. Build consistency. Always post about the same topics. Build trust. Build a nexus of people in that area the want the same kind of content. 7. The algo now includes negative signals like "not interested", "mute", "blocked". Rage baiting used to be good. Now it's terrible for your weighting. 8. Use formulas that are AB tested. When you find one that works, rinse it to get replies. Cringe, but it will really improve your weighting and reach. Note that everything you've learned about twitter before is now null and void. THIS is now the algo and it's based on Grok. Consider everything else algo based invalid going forward. TLDR: New algo. Write posts that make the right people stop, read, respond, share, and follow, while avoiding skips, mutes, blocks, and reports. Target your audience consistently.

Today I'm shipping Egghead. Andrej Karpathy's "LLM Knowledge Base" post in April nailed something: your AI should inhabit a knowledge base of files that you own, not locked in some opaque vector database. Notes (and memory) come first. I made a tool where the notes themselves can talk back. Egghead: markdown files in a folder on your computer, browsable in Obsidian, editable in vim, the whole thing. But any note can be more than a note. Add `class: agent` to its front matter and now the note is something you can chat with. A few of those in a folder and you have a working group: a researcher who knows your archive inside and out, an editor with strong opinions about your prose, a critic who's read everything you wrote last year and won't let you contradict yourself. The Karpathy pattern was "the AI maintains the wiki." Egghead's is "the wiki contains the AI." Agents can come and go but the knowledge base remains. egghead.computer

Not sure how anyone thinks GPT gives better code reviews than Claude. Running the same reviews side by side, hundreds of times recently, Opus 4.7 consistently outperforms GPT 5.5. This is in a wide variety of cases (front and back end).

Most launchpads solve the first 30 minutes of a token's life and walk away. You launch, then you're on your own. Blank is built for what happens after. And we're now LIVE! 🧵

It's April 2026 and we still need to add a "@AGENTS.md" line in our CLAUDE.md file. @claudeai, please fix this.

This is awesome. Give it a try! x.com/gwendall/status/203652…

This is really awesome. Everyone should try it.

Done a 180 on monorepos. I now love them. Sure the experience before wasn't great, but in smaller teams with more AI and first class support in Bun, things are very different today.

Anyone else getting ridiculous memory usage warnings in Claude Code? @claudeai - this is only happening in the latest version. Looks like a memory leak.

My information consumption is now 1/4 X, 1/4 podcast interviews of the smartest practitioners, 1/4 talking to the leading AI models, and 1/4 reading old books. The opportunity cost of anything else is far too high, and rising daily.

If you configure any environment with a collection of linux commands in 2026, you've failed. Any infrastructure should be defined in auditable code, be reviewable, replayable and if it fails, you start again. It's the only safe way to do it nowadays. Don't compromise.

The S in LLM stands for security.

Restart your Mac regularly Before restart: 5gb free After restart: 55gb free

Rumor has it that the following models are about to drop. Pretty exciting! GPT 5.3 Gemini 3.1 Deepseek v4

It's 2026. If your docs site can't be easily read by AI, what are you even doing?