Harvard Business Review · Article

Teach Your AI How You Make Decisions

Everyone has access to the same models. The companies scaling agents are the ones that wrote down how their best people decide.

Published
Reading time
About 12 min
Authors
Jen Stave, Ryan Kurt, John Winsor
Read the original

What is it about?

The authors argue that the bottleneck for AI agents is no longer the technology but an organization's ability to make its judgment explicit: risk tolerance, exception handling, escalation thresholds and brand voice that today live in experienced people's heads. Agents cannot absorb norms by observation. They act on what is written down.

Their answer is “judgment infrastructure”. Business units, HR and IT govern agents together as digital labor. Managers become “judgment architects” who codify their expertise, and employees learn to shape the agents acting on their behalf. The first codified use case is slow; later ones compound.

Key numbers

  • At ITA Group, the first six to seven months of codifying expertise were slow; afterwards, software timelines shrank from months to weeks.
  • A finance controller turned more than a dozen years of expertise into markdown files for his agents; a team of two now covers work that used to need ten people.
  • At Ramp, every employee learns during onboarding to build their own AI tools rather than act as a “button pusher”.

Why read it

  • It contains the most practical method in this library: a moderated panel of practitioners works through real edge cases. Where they agree fast, you have a policy. Where they disagree, you have judgment worth capturing.
  • It explains why agent pilots fail at scale even when the model is good.
  • It gives a CIO language for the org-design side of agents: who owns agent behavior, onboarding and retirement.

Who should read it

  • CIOs and AI leads moving agents from pilot to production
  • Operations and HR leaders who will co-own digital labor
  • Domain experts asked to “document their process” for an AI team

About the author: John Winsor

Executive fellow, HBS AI Institute; author of “Open Talent”

A serial founder whose career runs through magazines, advertising and open innovation. He mortgaged his house to buy the bankrupt Women's Sports & Fitness magazine, grew it into a media company, and later founded Victors & Spoils, which a Harvard Business School case describes as the first advertising agency built on crowdsourcing.

By his own account he set a speed record on Kilimanjaro in 1993 and won Harley-Davidson's business with a public tweet to its CMO. At Harvard he helped launch the Laboratory for Innovation Science before joining the institute behind this article.

Read it carefully: common misquotes

  • “Teach” here means written guidance and context files, not fine-tuning a model.
  • The cases are self-reported company stories, not a controlled study.
  • Jen Stave is the former inaugural director of what is now the HBS AI Institute; she moved to Keystone in May 2026.

How it shapes the CIO's AI Playbook

Notes by Pavel Veselov, last reviewed . Summaries are my own; quotes and figures belong to the original authors.

All 11 sources in the CIO's AI Library