March 2026

Pure software is rapidly becoming un-investable.

My first impressions with Codex (yeah I only started seriously using it this weekend): 1. It can find and fix bugs that Claude Code introduced (!) 2. It is not very good at UX/frontend - maybe there's a skill that can help? 3. $20/month seems like great value

I'd like to answer questions about our work with the DoW and our thinking over the past few days. Please AMA.

This paper is one of the first to test AI skills and the results seem to suggest that yes, they have high practical value. They use pretty mediocre skills (6.2/12 quality rating) harvested mostly from places like Github, and still get large boosts, especially outside software.

the real AI divide isn't technical vs non-technical anymore it's people who use AI agents daily vs people who don’t. And people building every day and people who don’t. the gap is getting enormous

Since Claud, OpenClaw, etc can hook up to any API, MCP, plugin, and more - I think it'll be extremely difficult for many other AI websites and apps to succeed. It's way easier to just ask your AI agent to do everything than have to navigate to separate websites or apps.

have you ever considered that your agent is lying to you and that it didn’t actually do what it said it did

I added a repo skill to Shiori called /pr-checklist that runs on (most) PRs. It double checks I'm not forgetting anything across like updating documentation (internal and external), testing, logging, schema changes, app versioning, etc. Super helpful! Here's an example: