Why forward-deployed engineering
A working model is not a running system. Forward-deployed engineering is the way to close the gap, and to leave it closed.
Writing
Notes on training engineers, forward-deployed work and putting AI into real workflows.
- Jev and System One models: a model that cannot generate text
A close reading of TypeSafe's Jev: a decision model with calibrated probabilities, and where it fits in enterprise AI.
- Harness engineering: the model plus everything around it
An agent is a model plus its harness. The harness is where reliability lives, and most of what you read about it is untested.
- The ontology is the product
Most enterprise AI failures are definitional. An ontology is how you fix that, and it can be smaller than you think.
- Spec-driven development: Spec Kit, AI-DLC and the discipline
Spec Kit and AI-DLC are two routes to the same discipline: write the work down before an agent builds it.
- How Reinforcement Learning Is Changing AI Intelligence
From human preferences to verifiable rewards, I examine what reinforcement learning changed and what remains uncertain.