Right now I'm deep into AI-native software development, rethinking how code gets written when AI agents do the mechanical work. The interesting question isn't whether AI can write code. It's what the job of a software engineer becomes when it can.
- Verification as the primary skill — if AI writes the code, building the system that catches when it's wrong is the actual craft. I wrote about this as The Dark Software Fabric: a 7-layer hierarchy from types to E2E tests that lets autonomous agents ship with confidence.
- The human role in AI-native workflows, aka The Ralph Loop — architecture, intent, and knowing what to verify.
- Why boring tech wins harder now — mainstream stacks compound better when your coding partner learned from millions of examples of them.
- Security of autonomous agents — prompt injection, containment, treating every AI-touched system as potentially compromised from day one.
- wt — git worktree manager with deterministic dev server ports
- trigger-cli — CLI for Trigger.dev: list, search, and run tasks from your terminal
I write longer pieces at jw.hn.