Peter Eysermans

The boring parts make AI agents production-ready

We have all been in the AI loop of hell, where the agent keeps going in circles to spoon-feed us the same knowledge base article for the third time. Support is an obvious use case for agents because the raw material is already there: documentation, policies, previous tickets, and known

Debugging should not become a therapy session

Debugging starts with reproduction. If you can reliably reproduce the issue, you are already halfway there. From there, you can form a hypothesis, test it, and work towards a fix. LLMs can be a real catalyst in that process. Most of us remember the first time we described a bug

If only AI can operate your product, you have already lost control

A risky pattern is emerging: teams build products with AI agents, then forget to build the boring screens, logs, and manual controls around them. Over time, they end up with systems they can no longer understand or operate without it. We already learned this lesson with vibe-coding: speed feels great

My First Personal AI Agent - J.A.R.V.I.S.

Since I started using Claude Code, I've wanted to build a truly autonomous agent: one that keeps working when I walk away from my desk.

Your engineering practice as an agentic pipeline

Last week we had a message from a user asking about a Safari bug. The bug has been lingering for a while. I've been experimenting with an agentic pipeline and decided it was time to test the next level. I gave it the Safari bug and it shipped

If the guardrail is a line in a prompt, the model can ignore it ... and eventually will

I learned this building an agent pipeline. An engineer agent would implement a feature, then a reviewer agent was supposed to check the work. More often than not, the review got skipped. So I added a line to the agent instructions: "All changes must be reviewed. Do not skip

Agent teams aren't there yet

Claude Code shipped a Teams feature. Give a group of specialised agents a task and let them tackle it together. An orchestrator coordinates, and sub-agents execute. On paper, this is the future of autonomous software development. In practice, it breaks down fast. I built a development pipeline with it. An

The cost of coming back to a piece of code has plummeted

We used to put a lot of effort into even the smallest code change. Coming back to it afterwards was expensive. Coding is no longer the limiting factor, human time is. Code reviews are where this plays out every day. They make sure the solution is sound, shared amongst the

A premium AI subscription costs almost nothing compared to what you're already paying that engineer

The first reaction most leaders have to AI is to plan. Figure out a strategy first, then act. The problem is that the tools and the landscape are changing so fast that by the time you've decided on an approach, it's already outdated. Companies have meetings

We went from vibe coding to agentic engineering. Next up: building the harness

We realised that vibe coding gave us quick results, but these were unmaintainable over time. So we swung to the other extreme: stepping through every line the agent delivers. Good quality, but you become the bottleneck. The next step is building the harness: the constraints, skills, hooks, and conventions that