What we learn operating AI

Every post draws on a system we engineered and still answer for in production, written by the engineers and leaders accountable for it, so each lesson has already been tested against real users and real auditors.

Define an Outcome

Agentic AI27 July 2026

Human-in-the-Loop Data Capture

Most capture pipelines put the reviewer at the end of the line, where a finished extraction looks settled and gets waved through. Oversight earns its keep at the point the agent is unsure, which makes this a placement decision before it is a staffing one.

Shardul SinghCTORead the blog

Tell us the number you need to move

Define an Outcome

What we learn operating AI

Human-in-the-Loop Data Capture

Are Frontier Models Becoming Commodities?

Measuring Agent ROI

Data Privacy and Residency for AI in Banking

Customer-Support Agents Compared

AI for Accreditation and Citizen Services

AI in Mental Health Support

A Risk Framework for Agentic Systems

GPT-5 in Production

Geospatial AI for Catastrophe and Property Risk

Durable Execution for Long-lived Agents

Fair and Auditable

On-Prem vs Cloud AI for Hospitals

Tell us the number you need to move