Surface Research
Notes from production
Notes on evaluating, benchmarking, and shipping enterprise agents. Sealed environments, calibrated judges, release gates, and what actually breaks in production.
The Ins and Outs of Simulated Worlds
Agents now do real work: refunds, records, automations, code. A bad version gets that work wrong on real customer data.
Surface Research·July 31, 2026·8 min read
How the Industry Is Designing Production Agents
The industry is settling on a way to design production agents.
Surface Research·July 26, 2026·7 min read
Dispatches From the Enterprise Agent Front Line
What we've learned from talking to dozens of teams building, buying, and running AI agents at enterprise scale.
Surface Research·July 17, 2026·10 min read
Efficient Agent Evals: From Production Traces to ROI
Building enterprise agents is not the hard part. The hard part is knowing, release after release, whether they got better.
Surface Research·July 9, 2026·13 min read
Offline Eval Environments: The Infrastructure Behind Agentic CI/CD
The infrastructure that makes offline evals actually work.
Surface Research·June 30, 2026·7 min read