AI-assisted development.
What actually changes when agents write most of the code. These posts cover the tooling I use daily, the failure modes I have hit, and the parts of the work that stubbornly still require a person.
What changes when agents write most of the code?
After thirty years of professional development I canceled my JetBrains subscription, not because IDEs got worse but because the work moved. Natural language, agent orchestration, and systems thinking are now the primary interface, and the expensive part of programming is deciding what to build and whether the result is any good. The Post-IDE Era makes that case, and From Shell Scripts to AI Organizations places it in the longer history of automation moving up the stack.
How do I use Claude Code on real production sites?
The demo is the easy part. AI-generated sites stall on the boring production work: prerendering, sitemaps, structured data, share images, analytics, deploy pipelines, and hreflang. I capture each of those once as a reusable Claude Code skill, with its gotchas written down, so the next site gets it right instead of relearning it. Reusable Claude Code skills for production websites walks through the approach. Claude Code’s habit of stating what it verified, assumed, and skipped is part of why I trust it with that work, as I argue in Your AI Coworker Might Be More Honest Than Your Human One.
What still needs a person?
Generating code got cheap; judging it did not. Unreviewed AI output handed to someone else is a slop grenade, and avoiding it takes experienced engineers, real review, and tools built for accountable work. The same cheapness makes scope creep worse, because every tweak is now affordable, and it makes taste the bottleneck: knowing what to make, what to cut, and when something is good enough to ship.
Where is this heading?
Agents are becoming users of the web in their own right, which forces a new trust model after decades of blocking bots. They may eventually want a programming language designed for them rather than for human readers.
How To Avoid “Slop Grenades”
Unreviewed AI output creates work for everyone downstream. Avoiding slop grenades is a management problem: senior engineers, real review, serious tools.
Read→The Post-IDE Era
Why I canceled JetBrains and moved beyond the IDE: natural language, agent orchestration, and systems thinking are becoming my primary programming interface.
Read→The Evolution of Automation: From Shell Scripts to AI Organizations
Automation keeps moving up the abstraction stack, from shell scripts and infrastructure-as-code to orchestrated AI organizations.
Read→Your AI Coworker Might Be More Honest Than Your Human One
Claude Code’s habit of explaining what it verified, assumed, and skipped can sound funny, but it may be a standard human coworkers should emulate.
Read→The Internet Is Becoming Agent-First: From Fighting Bots to Empowering AI Agents
The web spent decades blocking bots. AI agents now need access. The agent-first internet will trade friction for convenience and force a new trust model.
Read→What Will Be the First Agent-Native Programming Language?
Every programming language we use was designed for humans to read. As agents write more of the code, what does a language designed for them instead look like?
Read→Scope Creep in the Age of AI
AI makes every product tweak fast and satisfying. The real challenge is resisting endless polish long enough to protect the critical path and ship.
Read→When Everyone Can Build, Taste Is the Bottleneck
AI made building software almost free. The scarce input now is taste: knowing what to make, what to cut, and when something is good enough to ship.
Read→