2 min read
The demo was never the hard part
Pilots don't fail on model quality. They fail on the twelve systems between the demo and a finished piece of work — and on the exception queue nobody scoped.
- production-ai
- integration
- enterprise
Notes from the production side.
Short pieces on what breaks between a demo and production. Written by the people who fixed it, mostly after fixing it. No listicles.
2 min read
Write the evals before the prompt
An eval suite isn't QA you add at the end. It's the specification — and the only thing that lets a CXO answer 'is it working?' with a number instead of an opinion.
- evals
- reliability
- governance
2 min read
How much autonomy should an agent have?
The right answer is a configuration value, not a philosophy. Design the approval gates before the tools, start tight, and let the pass rate buy each loosening.
- agents
- governance
- risk
The posts argue it. The studies price it.
The same workflows from the other end: what the baseline was, what we built, and what moved once it was live.