Appiah AI
Insights
Notes on building AI systems people can trust.
Short, practical writing from Appiah AI on strategy, implementation, governance, and the messy middle between prototype and production.
The pilot is not the product
A pilot proves possibility. A product proves repeatability. The difference is ownership, monitoring, user behaviour, exception handling, and a clear route for improvement.
Retrieval quality is a management issue
Poor retrieval is often blamed on tooling, but the deeper causes are document sprawl, missing source ownership, weak metadata, and no cadence for knowledge maintenance.
Agents need boundaries before autonomy
Agentic workflows become useful when the system knows where to stop, when to ask, what evidence to cite, and which actions require human authority.
AI governance should feel operational
Good governance is not a PDF graveyard. It is a set of living controls embedded in intake, release, monitoring, procurement, and incident review.
The best metric is changed work
Accuracy matters, but the strongest proof is whether the system changes cycle time, quality, capacity, decision confidence, or customer experience.
Why minimalist AI strategy works
Clear constraints make AI projects more credible. A restrained portfolio lets leadership fund systems that are specific enough to test and important enough to sustain.