Resources
Writing on building AI that holds up under scrutiny.
Long-form pieces on the parts of enterprise AI that are genuinely difficult — grounding, evidence, evaluation and governance. Written from systems in production rather than from the discourse.
In progress
The model judges, the code scores
Where to draw the line between what a language model should decide and what belongs in ordinary deterministic arithmetic — and why getting it wrong makes a system unauditable.
Integrating the systems that do not want to be integrated with
A practical account of reaching enterprise platforms whose APIs are disabled by policy: what is legitimate, what is fragile, and what to build so it survives the first session expiry.
Governing what you do not host
Why software-as-a-service inverts the inspection model that enterprise standards are built on, and what a vendor standard has to contain instead.
Would rather just talk it through?
Most of what is written here started as a conversation with someone facing the problem for the first time.