AI

I had become the postman between my agents and the business, so I built a small MCP service that lets agents ask people directly and took myself out of the loop.

A self-hosted memory store I own end to end, so my agents' working context survives the loss of any account, vendor or machine, and nothing goes down a memory hole unless I put it there.

One small app on my Mac that turns Slack channels into Claude Code agents, with memory, feedback and lots of guardrails, and that I can talk to from the car through CarPlay.

Building a retrieval-augmented agent that scrapes a website with Firecrawl, indexes it in Qdrant, and answers questions about it using LangFlow — no custom orchestration code required.

AI-assisted app development can boost productivity, but it comes with challenges like inconsistent UI generation, unnecessary duplication, and prompt management issues. By organizing requirements in indexed files, generating standards documentation, and codifying rules with custom prompts, I streamlined my workflow and improved alignment across UI, functionality, and architecture.

Lessons from rebuilding a blog using AI tools — what worked, what didn’t, and why engineering fundamentals matter more than ever.