Blog
Essays on harness engineering and the human context layer — architecture, schema design, trust, and integration.
The missing layer in the agent stack
Harnesses standardized tools, memory, and evals for the machine half of the stack. The human half still enters as pasted prose. Career context is a supply problem — and supply problems need middleware.
Wire a human into your harness this afternoon
A practical walkthrough for builders: scope a key, read career context over REST or MCP, and decide when your agent should read — and when it should ask.
A schema, not a quiz
Self-report instruments are measurement schemas in disguise: versioned scoring, confidence as a first-class field, layers with separate audiences. The quiz is the interface — the schema is the product.
Human data needs harness discipline
Benchmarks earned trust through harness discipline: provenance for every number, reproducible methods, logged runs. Data about people deserves the same — consent in the schema, reads on the record, revocation that works.