Metacog

The problem I am working on

A professional can use an approved AI tool and still reach a conclusion whose basis is difficult to explain. Access controls say who may use a system. They do not, on their own, show which assumptions were challenged, what changed along the way, or who accepted responsibility for the result.

Metacog is my main research and development project. It investigates how that gap can be addressed through the design of the working process.

The central design decision

I separate AI capability from the authority to rely on its output. A model can help explore, draft and challenge; an identifiable person remains responsible for consequential professional judgement.

That distinction needs to survive the workflow. Exploratory material should not acquire the standing of reviewed advice simply because it has been copied into a polished document. The intended use of the work, its sources, the challenge it has received and the human decision to rely on it need to remain connected.

The architecture therefore brings together explicit working contexts, structured challenge, provenance and human responsibility. It is designed to sit across AI models rather than make an organisation’s governance depend on a single provider.

The commercial question

For a professional organisation, a useful AI workflow has to be both productive and reviewable. My design question is how to make the basis for reliance inspectable without imposing so much process that people work around the system.

Privacy is part of that question. More telemetry is not automatically better assurance. A governance record should not quietly become a mechanism for judging employees through behavioural monitoring. The architecture treats access, retention and the separation of content from governance evidence as substantive design choices.

What this establishes and what it does not

The governance model and system architecture have been developed through specification, challenge and refinement. Work is now focused on a demonstrable implementation and testing the assumptions against professional workflows.

This is not a claim that Metacog has solved regulatory compliance, guarantees correct advice, or has been validated in a live client deployment. The next test is whether the controls remain useful, proportionate and understandable in actual work.

What I can show today is how I translate a difficult accountability problem into system boundaries, workflow decisions and questions that an implementation can be tested against.