A brand system as a machine-readable spec
How this brand is built: rules written as a specification in version control, one generator, and every asset produced by a build rather than by hand.
Insights
Engineering notes from under the model
Notes on the work, written to be useful whether or not you ever contact us. No client data, no benchmark we did not measure, and no competitor named.
Published
Why AI pilots do not reach production
Four failures sit under the model, and a project hits them in order: unprovisioned data, infrastructure that cannot hold it, no security model for software that acts, and a legacy estate that cannot be paused.
An authority model for agents that call tools
An agent with a service account is an unbounded principal. What an authority model contains, who signs it, and why it has to exist before the agent does.
What is coming
- Data contracts before pipelines
- Budgeting capacity, latency and cost before the build
- Least privilege for software that acts
- Writing a rollback and cutover plan for an AI workflow
- How to load-test an LLM workflow — method, not results
- Evaluation harnesses for agents
- What open by default means in a contract
One a month, held. Two a month for three months and then nothing is worse than one a month forever.
Send us the problem
Describe what is stuck. We will come back with an architecture, a timeline and the first commit.