Insights
Field notes on shipping AI inside real systems.
Page 2
Automating invoice processing: the hard part isn't reading the invoice
Extraction is the easy half. The system that pays off is the one that knows which invoices a person should look at — and gets that list shorter every month.
Read post
A domain-expert agent is not a chatbot with your logo on it
Grounding, structure, evaluation, and a human gate: what turns a general model into an agent that knows your business — and can be trusted to act inside it.
Read post
AI can write the code. It can't be accountable for it.
Coding agents are fast and confidently wrong. The teams that ship well with them didn't find a better model — they kept the discipline the model can't supply: specs, review, tests, gates, and a name on every release.
Read post