What your team does with the time AI gives back is the tell.

Two teams face the same change: AI makes building faster.

Leaders in that position face a choice about what to do with the time it gives back. They can build more or learn more before committing. Building more risks hardening assumptions into commitments faster, at a cost that arrives later. Learning more risks moving slower now.

The first team chose to build more while learning less. Leadership prizes time to market, so the team cut back on discovery and doubled down on delivery. With little experience launching a product like this, it returned to familiar habits: an assembly line measured by how much ships and how quickly. Internal opinions stand in for customer evidence, and shipping features counts as progress.

The second team chose to learn more. It created a senior role to understand the needs of the people it serves and use that evidence to shape strategy before committing to what gets built. As it navigates a major product transition, it is strengthening the practices that help it escape what Melissa Perri calls the build trap, including a shared definition of the outcomes the team is working toward.

It is too early to assess either team’s results, but the difference is already visible.

The skill isn’t new

Much of the conversation about AI now says judgment is the new skill: knowing what deserves to be built, which trade-offs matter, and whether the result creates value. Judgment matters, but the skill is not new. Many organizations already employ people who practice it, in research, design, and product discovery, fields built on decades of human-centered design.

By 2017, Agrawal, Gans, and Goldfarb were arguing in Harvard Business Review that cheap AI prediction would make human judgment more valuable. Their book, Prediction Machines, made the same case, which a strategy+business review summarized in 2018. The authors define judgment as specifying the objective and weighing the payoffs of each possible outcome, which depends on knowing what people value. Human-centered design, rooted in industrial design and human factors, has provided methods for finding that out for decades. Henry Dreyfuss wrote Designing for People in 1955, and Don Norman’s The Design of Everyday Things (1988) brought the ideas to a wide audience.

The 2025 report from DORA, Google Cloud’s long-running research program on software delivery, describes AI as an amplifier of an organization’s existing strengths and weaknesses. AI exposes which teams were already doing the work.

Why leaders skip it

The cause is rarely ignorance. A leader can be criticized this quarter for pursuing evidence that could change the roadmap, but may never be blamed next year when the product misses the market. I have called this pattern the Normalization Law: the assumption becomes a requirement, and the requirement becomes a commitment.

Dan Lawyer, a former chief product officer at Lucid who now advises product executives, recently argued in a post that the bottleneck is not decision making but the learning that precedes it. Product teams already decide at an alarming pace to keep their builders fed. What has deteriorated is the hit rate, because those decisions lack learning rigor. Call that work diagnosis: learning about people and context before making a decision.

The pattern predates AI. McKinsey’s 2018 study of 300 listed companies found that only about half did user research before generating their first design ideas, and that fewer than 5 percent of respondents said their leaders could make objective design decisions.

The tell

One question separates the two teams: what does the team do with the time AI gives back? Three markers show the answer.

  • Diagnosis: is there a step to learn about people and context before work is committed to the roadmap?
  • Influence: can evidence still change the plan, and is there a point at which it can?
  • Accountability: are people accountable for outcomes, or for how much and how fast?

Learning (“let’s find the root cause first”): diagnosis before the decision, influence that can change the plan, accountability for outcomes. The team is using AI’s speed to learn.

Building (“we’ll test it later”): diagnosis after the decision, influence too late to change the plan, accountability for how much and how fast. The team is in a build trap, so AI’s speed builds more of what it already believed.

The three work together: outcomes without diagnosis and influence are just another target.

These questions describe how work is managed, not who is at fault.

Influence is where the second team’s approach differs most. What the diagnosis finds is meant to feed the work directly: the people building start from it, and findings can stop or reshape work before it is committed. That is how learning reduces risk: the wrong thing is caught while changing it is still cheap.

One quick check I’ve relied on across dozens of launches, as a consultant and in house, is the one I described in The Brown M&M Test: listen for what a leader means by UX. If they treat UX and UI as interchangeable, design is probably a production function in their mind, and the reaction when the difference is raised tells me more than the answer does.

What leaders change

Leaders who want better judgment need to let diagnosis, and the evidence it produces, change the plan. That means involving people with research and design expertise while the problem is still open, letting findings challenge a proposed solution, and evaluating what happens after release. Done early, this cuts risk before it becomes rework.

If the second team’s approach works, finance could see less rework and, over time, a faster payback on the transition.

Human-centered design provides the methods, and AI makes faster experiments and cheaper changes of direction possible. Whether teams use them depends on what leaders reward and what they allow teams to question.

Judgment isn’t the new skill.
Giving it influence may be the change organizations still need to make.

Sources