Insights

Articles by Jason Southgate

Articles Jason first published on LinkedIn, republished here in full. Each page shows the date of the original LinkedIn publication and links to it.

  1. Logs Are Not Evidence

    Originally published on LinkedIn on

    Logs show what an AI system did; evidence shows whether it was allowed to. What an evidence record must answer, why allowed is not the same as done, and how to keep proof without hoarding content.

  2. Shadow AI Is Customer Feedback

    Originally published on LinkedIn on

    When the approved AI route takes weeks and a personal account takes ninety seconds, people take the workaround. Shadow AI is a demand signal, and the answer is a paved road worth using, measured like a service.

  3. Model Gateway ≠ AI Control Plane

    Originally published on LinkedIn on

    A model gateway is a sensible first move, but it governs only the traffic in front of it. One coding agent's three operations show why the rules need an owner beyond the gateway.

  4. Who Actually Made This AI Request?

    Originally published on LinkedIn on

    Approving a model says nothing about the request. Governance needs to know who acted, on whose authority, with which data and through which route, and to record that as metadata rather than a prompt archive.

  5. Mind the chasm

    Originally published on LinkedIn on

    InfoQ's 2026 Cloud and DevOps trends chart puts AI gateways, MCP tooling and agents on one side of the chasm and mature platform components on the other. What is missing is the governance that joins them up.