The Log
What we are thinking about
Notes from actual project work — not what AI is going to do eventually, but what it does and does not do right now.
What AI agents actually do for operations teams
Not what the demos show. What happens after six months of daily use — the routines that stick, the places agents disappoint, and what that means for how to scope the next one.
The no-code automation ceiling: when Make stops being enough
A pattern we see repeatedly in teams that have been using Make or Zapier for a while. What the actual limits are and how to think about the transition.
Adding LLMs to an existing product without breaking what works
Practical notes from three integration projects. What we changed our minds about midway through, and why the scoping conversation matters more than the model choice.
When agents fail silently: four modes of quiet degradation
The failures that do not throw errors are the dangerous ones. What to watch for, why they happen, and how we catch them before clients do.
The EU AI Act for teams under 50 people: what actually applies
Most coverage treats the regulation as a large-company compliance problem. What it means in practice if you are a small team building or deploying agents inside the EU.