Building with AI

Answering the four framing questions prevents months wasted building something nobody asked for.

Climbing up is easy; ripping out an over-built chatbot is painful and public.

Each rung adds cost and complexity, so only climb when the simpler approach provably fails.

Confusing the two is the most common and expensive AI-building error teams make.

Large context windows handle a few docs without the overhead of a retrieval pipeline.

A concrete test set replaces opinions with an honest number telling you what to fix.

Five fields from twenty vendors ships this week; "all invoices" never ships at all.

Auditing representative, labeled, legally-usable data prevents demos that collapse on messy real inputs.

Tracking the score over time catches silent regressions before they reach production unnoticed.

Guarding keys like passwords stops accidental exposure when you share code.

Guardrails and human confirmation stop buggy or tricked agents from causing irreversible damage.

A capped, logged loop prevents runaway agents and shows exactly what actually happened.

Classification on a small model can be fifty times cheaper with negligible quality loss.

Swapping between Claude, GPT, or Pinecone later avoids a costly full rebuild.