Prototype vs. production AI

A prototype has one job: prove an idea is possible. It runs once, on data you chose, with a person watching. That is genuinely useful — it removes doubt before money is committed.

A production system has a different job: run every day, on data nobody curated, without supervision. That difference is where most of the engineering actually lives — validation, retries, failure states, logging, permissions, monitoring, handover documentation.

The gap is not a small final step. In my experience it is the majority of the work. This is why I show a proof of concept first, clearly labelled as one, and quote the production build separately.

Agree what a correct result looks like

For an enquiry-processing workflow, success is more specific than producing a plausible summary. The output needs the agreed fields, a traceable source and a clear destination. Missing details and ambiguous requests need an exception path. Test examples should include ordinary inputs as well as incomplete, repeated and unexpected ones.

A prototype can show the interface and the proposed handoff with illustrative data. The production scope establishes the supported integrations, access permissions and acceptance criteria before real records are changed.

Design for the step that cannot finish

If the next system is unavailable, the item needs a visible state and an owner. Retrying a failed operation must account for whether the first request already created a record. A useful workflow shows what completed, what is waiting and what a person must resolve.

Monitoring and recovery are operating requirements, not just features in a demonstration. The handover should explain how the team inspects a failure and who handles changes to the connected tools.

Keep the boundary visible

Healing Pott is labelled as a prototype with demo data in the work portfolio. Cadence Ops is described separately as a live operational product. Those labels matter: an interface preview does not demonstrate production data handling, user permissions or a measured business outcome.

When commissioning work, ask which parts of the demonstration are illustrative and which will be delivered in the production build. Agree the checks, deployment and support scope alongside the price.