A third of MLB dugouts were quietly running AI. The league banned it on purpose.

A third of MLB dugouts were quietly running AI on dugout iPads to call pitches and substitutions. This week the league banned it. Not for cheating. On purpose.

Here's the part that matters. No team broke a rule. There was no sign-stealing scandal, no hacked feed, no illegal earpiece. The tablets were league-issued. The AI worked. A June 11 memo took effect July 15 for the second half of the season, and the league drew the line anyway, after a review found as many as a third of teams reaching generative models mid-game for strategy calls.

Read that again. The tool worked and they banned it.

Every AI story you read is about failure or fraud. The model hallucinated. The chatbot lied. Someone cheated. This one is different. The league looked at a tool that performed, that gave real answers, that no rule forbade, and decided the in-game call belongs to a human. On purpose. By design.

That's the read most people will miss.

The line is the strategy

Most operators are running the opposite play. The pressure is one word: everywhere. Put AI in every workflow. Add a copilot to every screen. Someone runs a "where can we add AI" sprint and the score is the count of places you added it.

That's a bad game. It treats every decision as the same kind of decision. It's not.

Some decisions are worth more because a person visibly owns them. Who to fire. What to bid on the deal. Which customer to walk away from. The value isn't in the answer. The value is in the accountable human standing behind the answer, whose name is on it, who can be looked in the eye. Push that to a model and you don't save time. You dissolve the trust that made the decision worth anything.

Baseball understands this in its bones. The manager walks to the mound. The stadium watches. If he pulls the pitcher and it blows up, that's his call and his career. Hand that to an iPad and the game is still played, but the thing people came to watch is gone. The judgment. The nerve. The person who owns the outcome.

Your business has mound visits too. You just haven't named them.

AI works. That's not the question.

Let me concede the whole skeptic's case up front, because I don't need it. AI works. The MLB tools worked. Foltrigg drafts filings faster than a first-year associate. Cosmos OS reads a deal room in an afternoon. The technology is real and it's here.

So the edge stops being "can we use it." Everyone can use it. The edge is knowing where you won't.

A no-AI zone is a positioning statement your customers can feel. When a client knows a person owns the call that affects them, they trust the whole operation more, and they'll pay for that trust, and they'll stay through the quarter where a competitor's fully-automated shop ships a cold, wrong answer at scale. The line you refuse to cross is a product feature. Say it out loud and it sells.

Withholding is a decision, and most operators never make it. They add. They never subtract. They optimize for coverage instead of judgment.

The scoreboard for AI is not how many places you put it. It's whether you knew which places to leave alone.

Run it backward before the offsite

Here's the fifteen-minute version. Do it before your next AI push, not after.

Name the three decisions in your shop where a customer's or an employee's trust depends on a human being accountable. Not the routine three. The heavy three. The ones where someone's livelihood or your reputation rides on who made the call.

Write them down. Put them explicitly off-limits. Tell your team those three are human-owned by policy, and say why, so nobody quietly automates them to hit a metric next quarter.

Now go automate everything else without guilt.

The teams that win the next decade won't be the ones who put AI in the most places. They'll be the ones who knew where the mound was.

Draw the line first. The tools come after.