Writing

AI steepened the performance curve instead of flattening it. Output used to tell you who the barrels were. It does not anymore.

Keith Rabois's framework on operators is more relevant than ever. AI just made ammunition dramatically cheaper. Barrels did not get cheaper at all.

The experienced users have a 10% higher success rate. And the gap is widening, not closing.

When AI generates the code, your job shifts from writing to verifying. Here is a practical playbook for the new reality.

The amount of time spent reviewing a pull request is inversely proportional to how much it can actually break production.

Sonnet + Opus advisor scores higher than Sonnet alone and costs 12% less per task. We were building this ourselves. Now we do not have to.

Anthropic studied 10,000 real conversations and built an AI Fluency Index. 85.7% of productive AI conversations involve iteration, not acceptance.

Most engineers are solving the wrong problem with AI. The real leverage is not getting faster, it is building systems that make asking unnecessary.

The real shift is not chat. It is agents. Drop an .agent.md file into your repo and stop repeating context.

Moltbook is literally Facebook for AI agents. I genuinely could not tell which posts were autonomous agent musings versus human-prompted content.

The architecture is worth studying. The security model is not ready for production. Both of these things can be true.

Have not had this 'wait, this changes everything' feeling since ChatGPT dropped. It is what Siri was supposed to be 13 years ago.

The research is real but the headline is incomplete. Here is what the study actually found, what it missed, and what to do about it.

Anthropic just published an excellent deep-dive and it crystallizes something I have been seeing across enterprise AI deployments. Prompt engineering was act one.

Why AI tools deliver 300% gains for some teams and 10% for others. The answer is not the tools.

The difference between mediocre and exceptional AI output is not the model. It is the prompt. Here is how to treat prompts as engineered artifacts.

After analyzing hundreds of AI-generated code issues, six categories of missing context explain nearly every failure. Here is how to fix each one.

The gap between AI-assisted teams (10-30% gains) and AI-ready teams (100-300% gains) is not the tools. It is the context.

Individual agent intelligence matters less than seamless coordination. Three production-tested patterns for making agents work together.

When your agent workflow fails at 2 AM and you need to ship by morning, you learn fast.

Is it still coding if a significant part of your code is written by AI? The role of the staff and principal engineer is fundamentally changing.