Notes from the frontier of agentic AI development. Real experience, real insights, no hype.
A client who hasn't written code in years sent me two feature requests and a plan he'd gotten from Claude. The plan was good, and it was written blind. So instead of writing the ticket myself, I set him up to write it inside the repo, with the process loaded. Here's what happened.
Read more →I gave my agentic development process a rule borrowed from the Ritz-Carlton. Then I went to bed with ten tickets and woke up with twenty. A story about deferred debt, unintended consequences, and why 'later' means something different when an agent is running the show.
Read more →Seven weeks designing a way to run orchestrators in parallel. Three cold reads, each ending with more questions than the last. Then a go-around, and a two-day rebuild of the simple version Claude had been suggesting all along.
Read more →In March I described version 3 of my agentic development framework. Version 6 just shipped. The phases haven't moved; what changed is that every failure the process had became a numbered rule. The progression, version by version, and why each one happened.
Read more →My AI built three API integrations that looked professional, passed review, and were completely fabricated. What happens when the model doesn't know it's wrong — and why your experience matters more now than ever.
Read more →I upgraded from Opus 4.6 to 4.7 without thinking twice. It was like talking to a completely different person. A field report on what happens when you assume a new model version will just work.
Read more →Almost everything I've written has Claude in the driver's seat. But Claude can be just as useful in a support role — the expert on the phone while you're standing under the hood.
Read more →A question from the room at a live demo opened up a gap in my agentic pipeline I'd walked past for weeks. Advisories work because a human reads them. Pull the human out and they don't get ignored — they get unread.
Read more →My business partner wanted AI driving the workflow end to end. I wanted to build it for a human first, then hand it to the agent. The reason isn't philosophical. It's about what the model actually needs to do its job.
Read more →Your AI assistant thinks everything you do is brilliant. That's not collaboration. It's a participation trophy. And if you don't recognize it, you end up announcing table stakes like it's a revolution.
Read more →Computers don't multitask. They context-switch. Humans are even worse at it. Agentic development doesn't solve this problem. It amplifies it.
Read more →Claude confidently suggested a command that didn't exist. It looked right, it felt right, and it was completely made up. But hype is hype, whether positive or negative — and if you know to check, you can catch it.
Read more →In traditional development, everyone assumes certain things are "easy" — documentation, admin screens, user management. They never are. Agentic development is the first time the assumption actually matches reality.
Read more →The interaction with AI feels so natural you forget it's not human. Until the context window resets and your co-worker forgets everything. A field note on anthropomorphizing the tools we build with.
Read more →Standing in line for Pirates of the Caribbean, I decided to add an MCP server to an app I'd just built. By the time I landed, it was deployed. Here's why every API should have one.
Read more →A colleague had a product idea he'd been thinking about for a while. We spent an hour on the phone on a Friday. By Sunday night, it was built, tested, deployed, and documented. That's not a typo.
Read more →I decided mid-project to add comments to the entire codebase. Claude gave me three options, I picked one, and it went through everything without a word of protest. Try that with a team of engineers.
Read more →I handed AI-generated code to a senior engineer for a real code review. The line-by-line quality was solid. The architectural lessons changed how I build everything going forward.
Read more →I needed to test SMTP edge cases that no sane manager would budget for. Agentic development didn't just make it cheaper — it made it thinkable.
Read more →I expected Claude Code to tell me what changes to make to AgentZula's agent prompt. Instead, it realized it had MCP access to make the changes itself — and asked permission.
Read more →A detailed walkthrough of the five-phase agentic development framework I've built — from recording a conversation through orchestrated development with review gates, milestone checkpoints, and drift prevention.
Read more →How I went from using AI as a code helper to building a full agentic development framework — and collapsed a month-long project into a week.
Read more →Every company name has a story. Ours started with a line from a film, a career spent proving things could be done, and a philosophy that became an identity.
Read more →