Buy the AI license, keep the judgment

You bought the tools. You pushed the mandate. And the returns are thin. You are not alone, and it is not a technology problem. Let me catch you up on what changed in how the sharp teams are thinking about AI and their people.

More of the same, just faster

Here is the uncomfortable part. AI mostly amplifies how you already work. Some of that is useful. The rest is your bad habits, running at speed. Jeff Gothelf lays out the numbers: enterprise AI deployment jumped about 400% across 2024 and 2025, yet real returns sit in only 12 to 18 percent of companies. MIT's research found roughly 95% of GenAI pilots produced no measurable business value.

Gothelf pulls a story from Barry O'Reilly's venture studio that lands it. After ChatGPT arrived, founder pitches went from handcrafted decks to "data-room dream" packages where every artifact was present and polished. One or two follow-up questions showed almost no thinking underneath. The tool had absorbed the founder's judgment instead of sharpening it.

Spend thirty dollars a license across a thousand people and you can make your worst workflows faster. That is the trap most rollouts walk into.

Traits, then tasks, then tools

O'Reilly's fix is an order-of-operations move he calls the Trait-Task-Tool model. Most people run it backwards. They pick Claude or ChatGPT first, hunt for a task to point it at, and never ask how they actually do their best work.

Flip it. Start with your traits, the way you naturally perform. Find the high-leverage tasks where your judgment creates value. Then pick a tool that amplifies both. O'Reilly, who is dyslexic, spent years failing to write a book by typing. His real trait was talking. So he got interviewed, recorded it, transcribed it with AI, and edited that into chapters. His pace went from one chapter a week to about four hours a chapter. The tool did not save him. Redesigning the work around how he thinks did.

At the team level, the tasks worth redesigning are the recurring decisions, the go or no-go call, the prioritization meeting, the weekly review. People are paid to make decisions, not to generate output. Point AI at the judgment, not the busywork.

Resistance is a signal, not a bug

When your team drags its feet on the AI mandate, that is often a rational read, not fear. John Cutler makes this the whole ballgame with a simple formula: success equals Machine Understanding times Problem Understanding times Practice Evolution, all divided by the Social Contract.

The denominator is the part leaders skip. It is the promise that gains from improvement get shared with the people who found them, not used against them. Cutler is blunt: "Kaizen only works when there's a credible social contract that the gains from improvement will not be captured against the improvers." Break that trust and the whole formula collapses, no matter how good your tools are.

And because it is multiplication, any zero on top kills the result. "Everyone must use AI" is a practice mandate that assumes the other two terms take care of themselves. They will not.

The judgment you cannot download

AI floods the market with output that looks like work. That is exactly why real judgment gets easier to spot. Chandan Rajbongshi makes the case for designers, but it holds for anyone on your team. A certificate teaches process. It cannot teach the ability to look at a product and know in ten minutes what is broken and why.

His test is a good one to steal for hiring and reviews. Give someone a product they have never seen and ten minutes. Process knowledge runs a heuristic checklist and finds real things. Real judgment tells you the actual problem, why the team made the call that caused it, what constraint they were under, and what breaks if you fix it. That second answer comes from being in rooms where hard calls got made under pressure.

Same message from Netflix. Elizabeth Stone, Netflix's CPTO, now names systems thinking as the top skill she looks for, and treats AI fluency as a baseline expectation, not a senior perk. The tool is assumed. The thinking is the differentiator.

The deep cut

Governance is what protects the judgment when the pressure hits. Eric Ries points to Claude Code's shipping speed and says the velocity comes from a structure that lets the team invest long-term without quarterly pressure to cut corners. Ries warns about "financial gravity," the Costco lesson where one small compromise gets baked into the forecast and you have to do it again next quarter.

For you, the practical version is smaller and doable this week. Pick one recurring workflow that matters. Name what "good" looks like and what people do differently when it goes well. Write the new behavior down as a hypothesis. Only then bring in the tool. Run it for a short window, hold a retro, and kill, pivot, or persevere. Treat culture change like product development, one small experiment at a time. The teams that win the next two years are the ones that looked honestly at how they work before deciding what to automate.

Three questions for your team

1. Pick one recurring decision meeting. What would it look like if it went well, and which specific behavior could a tool amplify without replacing the call itself?
2. Where in your AI rollout have you promised people that the gains they find will benefit them, not be used to cut them? If you cannot point to it, that is your denominator problem.
3. In your next portfolio or hiring review, can your best people pass the ten-minute test, and is your team marketing that depth or hiding behind the same polished output as everyone else?