Founder & principal
Engineer
Name and bio coming soon. Previously built and shipped agent systems at large AI labs; spent the last few years closer to production than to research.
Core Automation is a consultancy of senior engineers. We help organizations adopt AI the most efficient, repeatable way we know — by engineering the systems that turn capable models into consistent quality results.
We're small on purpose. The kind of small where the person designing your harness is the person who built the last one. Where the proposal you read is the proposal we deliver against. Where the memo at the end of an engagement is short, because the work is in the repo — and so is the playbook your team uses next quarter.
Models are commodities. What makes an AI product feel reliable is everything around the model — routing, retrieval, refusal, recovery. The purpose of the harness is to get the same right answer twice. That's where the work is.
If your team can't agree on whether a change made the AI better, you're not shipping — you're guessing. The eval suite is how AI quality stops being a vibe and starts being a number anyone on your team can defend.
"We'll figure it out" is how six-month projects become eighteen-month projects. We write the spec, we agree on what done looks like, then we build.
A flashy demo that works once is theatre. A boring system that works the same way every Tuesday is a business. We engineer for the second one.
If we can't help you, we'll tell you on the call. If we think the architecture you've described won't work, we'll say so before you sign. The intro call is a real call.
The agent systems that survive in production look surprisingly like good distributed systems. Caching, retries, timeouts, queues. The novel parts get the headlines; the boring parts pay the rent.
Most AI projects die in the gap between a working demo and a system you can rely on. Closing that gap is the work.
A small group of engineers who've built agent systems at the kind of companies you've heard of, and shipped them past the demo stage.
Name and bio coming soon. Previously built and shipped agent systems at large AI labs; spent the last few years closer to production than to research.
To be filled in. Strong on retrieval, eval design, and the kind of CI plumbing that doesn't make the highlight reel but holds the system together.
For specialised domains — safety, formal evaluation, healthcare, regulated industries — we bring in advisors who know the territory. Always disclosed up front.
// Placeholder copy — drop in real bios, photos, and links when ready.
Selected pieces. We write rarely and only when we have something concrete to add. Subscribing is fine; nothing in your inbox more than monthly.
Why your eval suite is the only honest document about what your agent does. And why writing it last is a mistake.
Read → 8 minWhen the safest thing your agent can do is decline. A pattern we've reached for repeatedly in regulated domains.
Read → 6 minMost agent demos cluster at "pretty good." What it takes to push past it — and what stops most teams from doing so.
Read → 11 min