Codifying Judgment: The Real Bottleneck in AI Agent Adoption

The New Constraint in AI Transformation

Some organizations are using AI agents to fundamentally transform operations. Others remain stuck running small, low-stakes experiments that never scale. The difference isn't access to technology—most companies use the same models and infrastructure. The divide comes down to something most leaders have never confronted: making judgment explicit.

For decades, organizations never needed to articulate how their best people made decisions. Expertise transferred through mentorship and observation. New employees absorbed norms by watching experienced colleagues navigate ambiguous situations. That model worked when humans executed all the work.

AI agents have changed the equation. Unlike traditional software, they can operate in ambiguous environments and make real-time decisions. But unlike people, they can't absorb organizational culture through observation or infer unstated context. They operate based solely on what's made explicit.

This creates a specific failure mode now appearing across industries: Companies deploy customer-facing AI agents without first codifying how their best service representatives handle pricing exceptions, frustrated long-term customers, or requests that fall outside standard policy. The agent eventually goes off track because no one documented how those decisions actually get made.

Why it matters

The organizations that build "judgment infrastructure"—structured guidance that translates tacit decision-making principles into executable form—will gain compounding advantages in speed, consistency, and capacity for innovation. This represents the next competitive moat in AI adoption, far beyond simple access to models.

Three Structural Shifts

Organizations pulling ahead are making three fundamental changes, according to research first reported by Harvard Business Review.

First, business units, HR, and IT are governing together. Defining risk boundaries and performance expectations for agents are organizational questions, not purely technical ones. ITA Group, a global events and recognition company, learned this when building an AI agent for air-travel booking. The challenge wasn't building the agent—it was defining when to optimize for cost versus traveler experience and which exceptions were acceptable. The company responded by giving developers, managers, and knowledge workers tools to shape agents acting on their behalf.

Second, managers are becoming judgment architects. Debbie Riazzi, director of compliance at AWP Safety, built a portfolio of agents that each codify different slices of her expertise. One handles medical accommodation requests by pulling relevant job descriptions and surfacing how comparable requests were resolved. Another manages information requests by parsing what's being asked and routing to the right owner. The agents save hundreds of hours annually, but more importantly, they apply her judgment consistently across every case.

Nathan Mapp, a controller at a global venture-capital firm, codified his expertise into markdown files that his agents reference in real time. A team of two now covers ground that previously required ten people, with top-tier judgment applied to every detail.

Third, the "thought-doer" is becoming the most valuable employee profile. The traditional divide between strategic thinkers and operational doers is collapsing. High performers now reason strategically and operationalize their thinking through agents. Ramp, a financial platform serving 30,000 companies, equips every employee with ChatGPT Enterprise, Notion, and Perplexity during onboarding, training them to build their own AI tools rather than act as button pushers.

A Better Starting Method

Most organizations ask experienced people to write down what they know. That rarely works because experts struggle to articulate tacit knowledge in the abstract.

A more effective approach: Convene a small panel of experienced practitioners and walk them through realistic scenarios and actual edge cases. Where the panel agrees quickly, you have clear policy. Where they disagree, you have judgment worth capturing. The transcript becomes your first draft of codified judgment.

ITA Group's trajectory illustrates the compounding effect. The first six to seven months were slow as the company learned to translate expertise into context files agents could use. Once that operating model took hold, the pace changed dramatically. Developers began using agents to move from idea to working prototype in weeks instead of months.

When judgment is successfully codified, expertise becomes portable. Best practices are no longer locked inside senior people. Institutional knowledge deploys across functions, geographies, and products at scale.

The first phase of AI adoption was about access to the best models. That phase is over because access has been commoditized. The next phase will be defined by who has done the harder work of encoding how they actually think and work, as detailed in the Harvard Business Review analysis by Jen Stave, Ryan Kurt, and John Winsor.