I have watched this relief show up at more than one portfolio company this year. The inbox goes quiet, the pings slow down, and a leader reads it as proof the team learned to work alone.

Then a server goes down at two in the morning, and two of the strongest engineers spend an hour fixing the same incident from opposite ends, each sure they are alone on it.

Neither one has asked the other anything in weeks. Only the AI tool.

I have watched this pattern show up across more than one portfolio company this year. A team rolls out an AI assistant, usage climbs, and the volume of peer questions drops first: the quick "have you seen this before" that used to go to a colleague now goes to a chat window instead.

Managers notice the drop in interruptions before they notice anything else, and they read it as the team maturing (fewer dumb questions, more independent output, a calendar with more room in it).

What actually happened is narrower and less flattering. The person asking swapped a colleague who might push back, or ask a question of their own, for a tool that always answers and never remembers who asked twice.

It surfaces during a crisis, long before any dashboard catches it.

A 1,000-person study from BetterUp, run with the Stanford Social Media Lab, found employees increasingly go to AI for input instead of a colleague (the kind of shift an engagement survey never asks about). The shift measurably lowers team coordination and raises retention risk.

Employees who still go to a colleague first are 7.3 times more likely to report strong coordination and 9.5 times more likely to report strong team trust than employees who default to AI, a gap that shows up whether the employee is junior or senior. For a 10,000-person company, the resulting turnover alone runs about $27 million a year.

HiBob’s 2026 survey of 1,200 business leaders found 67% now tie AI skill to promotion and half tie it to performance ratings, yet only 36% see managers as prepared to coach that skill on a team.

Microsoft’s own research on 1,800 employees found a 30-point jump in trust when a manager visibly used AI alongside the team, instead of leaving people to figure it out alone.

This is not a buyout-portfolio problem. Any leader who rolled out a shared AI tool in the past year is running the same experiment, whether you operate three portfolio companies or run one 40-person agency.

The question was never whether adoption happened. It is whether your people still ask each other before they ask the tool.

The leader who pushed the rollout is the one who pays for it later, usually with interest.

The bill lands as a stalled incident call, a missed handoff, or a new hire who cannot find anyone willing to walk them through undocumented context, because the two people who held it (one now in a different department) stopped checking with each other months ago.

By then, the decision that caused it is impossible to name.

Nobody voted to stop checking in with each other. Each person just took the faster option, one question at a time, and the people who used to catch problems before they became incidents went quiet instead, with no meeting about it. The board sees the AI line item and calls it efficiency.

The leader sees the outage and knows better, with no clean way to explain the difference between a team that got faster and a team that stopped talking to each other.

Before the next AI rollout, or an audit of the one already running, build three concrete checks into how you track adoption, not just usage logs.

None of this slows the rollout down. It keeps a question going to a person often enough that the relationship survives the tool, which is the part no adoption dashboard measures.

A quieter inbox was never proof the team got better at its job. It only found a faster place to ask.

But the quiet you were enjoying was the warning.