The work behind the advice
Four strands, each marked with its actual state. One has a published method and is still collecting evidence, two are running, and one is a list of problems I do not have good answers to either.
Agent governance readiness benchmark
What organizations actually score across entitlements, enforcement, attestation and model risk.
Running · See the benchmark method
Open-weight parity evaluation
Whether a model you control can match a hosted one on the specific workflow, measured rather than argued.
Running
Agentic incident patterns
What actually went wrong in public agent incidents, reduced to the failure that carried it.
Running · First teardown: the Hugging Face credential
Open problems in agent governance
The questions this practice does not have good answers to, published deliberately.
Open question
Published reasoning is the proof
This practice names no clients and publishes no case studies, so the evidence that the advice is grounded has to be the work itself. Everything here is built from public sources, anonymous data or first-hand method — and the parts that are unresolved are published as unresolved.
If you work on any of this — particularly the open problems — the practice is glad to compare notes under NDA or in the open. Get in touch.
Bring the problem you cannot evidence
Research is upstream of the advisory work, not separate from it. If one of these strands maps onto something live in your estate, that is the conversation worth having.