Short, opinionated notes on automation, AI and the unglamorous work that makes them hold up in production.
All writing
Deterministic data beats AI interpretation
Pointing AI at a hard problem feels like progress. Often the reliable answer is already in the data, and the real skill is knowing the half of the problem AI shouldn’t touch.
ReadAI should never touch your data directly
Let a model query your database in its own words and it will hallucinate, miss fields, and build shadow copies of the truth. The fix is a layer, not a better prompt.
ReadHard requirements belong in code, not in the model
Every screening problem splits into must-haves and matters of degree. Hand the must-haves to code, the judgement to the model, and keep evidence under everything.
ReadReplace the paper, not the ERP
The floor doesn’t run on the ERP. It runs on whiteboards, paper slips and phone calls beside it. You don’t need a new system of record, you need to stop working in the margins.
Read“Review and approve” is the design, not a fallback
Human-in-the-loop isn’t a safety net you remove once the model is trusted. The approval step is the design, and it matters most where mistakes are expensive.
ReadMost automation projects fail before automation starts.
You can't automate a workflow that three people describe three different ways. Map the process first. Then talk about tools.
ReadWhy we don't sell projects.
Project work freezes the moment it ships. Operations don't. The retainer is how the operational layer stays clean as the business changes.
ReadAI is useful when it removes work.
AI extracts the data. Rules decide what happens next. Systems validate the result. Everything else is a demo.
Read