Insights
What we learn by creating and operating agents.
Insights into AI model economics, agent governance, agentic software engineering, and the Brazilian AI market. With sources for you to go deeper.
Subscribe via RSS- Models & Economics
Companies that don't measure or attribute cost per task to the areas using them are not operating AI agents — they're just paying for them
AI token prices fell 67% in a year, yet the cost of running AI agents went up. The explanation lies in call volume — and without attribution data, nobody can control it.
Read - Governance & Evidence
Evidence assembled after the fact is justification, not control
The difference between audit records generated by the operation in real time and records reconstructed on demand is obvious to an auditor — and by then it is too late.
Read - Agentic Engineering
AI performance moved: out of the models and into the execution harness
Holding the AI model constant and varying the agent harness: 58.0% versus 79.8% accuracy on Terminal-Bench 2.0.
Read
AI Strategic Roadmap
Where to start with AI, and what comes next.
Seven axes and 63 activities, from the first strategy document to steady-state operation. A map to place your company and choose the next step.