Notes contributor
Gopal Patel
Takes one workflow to production AI.
Founder and engineer. Chief technology officer at Auriens, where he built its early-stage ventures arm. He has structured cross-border transactions, worked in commodity broking, and created the by-appointment technology concierge service at Selfridges.
Meet Gopal on the team pageNotes from the work.
- Evals, runbook and handover: what must ship with an AI workflow
Evals give bounded evidence about behaviour before and after change. The runbook turns known signals and failure states into action. Handover gives the receiving team the artefacts, access, judgement and authority to operate both.
- How to choose the first AI workflow worth building
Choose the smallest consequential workflow whose need, system path, evidence, failure boundary, advantage and owner can be described before implementation. Pause when any of the four disqualifiers remains unresolved.
- What production actually means for an AI workflow
An AI workflow is in production when a bounded piece of real work crosses the organisation’s actual system path, has relevant evidence and explicit failure responses, is owned through change, and remains operable after the delivery team leaves.