AI & Tech
Every piece below is written by someone on our team who shipped what they're describing. No marketing edits. No ghostwriters.
Featured
A pattern we've seen at four separate clients: agents reach 60% accuracy in a week, then plateau for months. The cause is almost never the model. We unpack the four most common stalls and the playbook for each.
Read the field report →Recent writing
Treating evaluation as the spec — not the QA step — changes how teams build agents. Here's the architecture we've standardized on.
October 2026
Eighteen months of operations data from a 3M-member health plan. Where the agents excelled, where they failed, and how we adapted.
September 2026
A reproducible benchmark across enterprise tool-use scenarios — JSON conformance, error recovery, retrieval-grounded answering. Numbers, not vibes.
August 2026
SR 11-7 wasn't written for agents, but it doesn't need to be. We share the policy template our financial services clients have adapted.
August 2026
Lessons from running 47 production agent workflows across our managed operations practice. What pages, what doesn't, and how we tune.
July 2026
A two-week protocol for understanding an existing workflow before designing the agent that will replace parts of it. Templates included.
June 2026
Subscribe
A single email per month, summarizing the most useful piece we published and the most important thing we learned in the field. No promotional content, ever.
We sometimes co-publish with clients on engagements that produced unusually clean lessons.
Get in touch →