Blog

Notes from
the lab.

Writing on automation, applied AI, and how we build.

Automation that touches money has one rule: it must never fail quietly. The habits we picked up making invoice pipelines boring enough to trust.

Most retrieval problems are data problems wearing a trench coat. What actually moves the needle when you connect an LLM to company documents.

Not every manual process deserves software. The four questions we ask before we agree to build anything.

Big automation projects fail in familiar ways. Small, sharp tools survive contact with reality. On the thinking behind our first product decisions.

Put this thinking
to work.

If something here sounds like your problem, tell us about it.