AI Is Making Teams Faster, and Harder to Manage

Picture a Tuesday review on a 9-person product team. One engineer opens a pull request that would have taken a week last year. She built most of it in an afternoon with an AI coding assistant. Her reviewer, who writes everything by hand and is proud of it, leaves forty comments and a note that says “did you actually test this?”

Across the team, the designer uses AI for early concepts and doesn’t mention it. The product manager pastes AI-written specs straight into the tracker. One senior engineer won’t touch any of it.

Everyone is doing reasonable work. The team is still quietly falling apart.

AI made each person faster. It made the team harder to run.

Your Team Isn’t Adopting AI. Each Person Is.

Most AI rollouts are planned as if a team adopts a tool together. That’s not what happens. People adopt at their own pace, in private, based on their own comfort and curiosity.

The numbers show how wide the spread is. Gallup’s ongoing Indicator: Artificial Intelligence, which polls U.S. employees through its probability-based Gallup Panel, found that as of May 2026, 52% use AI at work a few times a year or more, 30% a few times a week or more, and 15% daily. Only 25% said their organization has communicated a clear plan for integrating AI into current practices.

Put those together and a small team could look something like this: a couple of people who use AI for almost everything, a few who use it now and then, and some who barely use it or have decided not to.

Microsoft, which sells AI tools for work, sees the same gap. Its 2026 Work Trend Index, published in May 2026 and based on a survey of 20,000 knowledge workers who use AI at work in 10 countries, puts it plainly: “In many cases, employees are moving faster than the organizations around them.” Only 19% of those AI users landed in the group Microsoft calls Frontier, where both the individual and the organization are ready.

The Real Problem: Nobody Can Read Effort Anymore

Here is what usually gets missed. Teams have always relied on an unspoken sense of how much work something took. A long document meant someone thought hard. A quick reply meant it was a small ask. A big pull request meant days of effort, so you reviewed it with respect.

AI breaks that signal. A 12-page spec can now take ten minutes. A polished deck can hide the fact that nobody checked the numbers. And the people who don’t use AI read the output with the old assumptions, so they either over-trust it or suspect everything.

We think that is where a lot of the friction comes from. It’s not that heavy users and light users disagree about AI. It’s that they no longer agree on what “done” looks like, how long things should take, or how carefully something needs to be reviewed.

Why It Stays Hidden

You would think these differences would come up in conversation. Often they don’t.

In the Slack Workforce Index, published in November 2024 by Slack’s parent company Salesforce, which sells AI products, nearly half (48%) of desk workers said they were uncomfortable admitting to their manager that they had used AI for common tasks. That survey of 17,372 workers in 15 countries was fielded in August 2024, so it’s worth reading as a snapshot from an earlier stage of adoption. The top reasons: it felt like cheating, and people worried about looking less competent or lazy.

That fear isn’t irrational. In September 2025, BetterUp, a coaching company, and the Stanford Social Media Lab surveyed 1,004 full-time U.S. desk workers about workslop: AI-generated work that looks polished but is unhelpful or off the mark. Forty percent said they had received some in the past month. On average they spent close to two hours dealing with each instance, and about half said they saw the sender as less capable and less reliable afterward.

So the heavy users hide their use, the light users assume the worst, and the manager sees neither.

Five Ways to Manage a Team That Uses AI Unevenly

You don’t need a 30-page AI policy. You need a few shared rules that make the invisible differences visible and fair.

1. Agree on norms by type of work, not by person

Sit down as a team and sort your work into three buckets: where AI is encouraged (first drafts, research summaries, test scaffolding), where it’s fine with a second look (customer emails, specs, code that touches payments), and where it stays out (anything confidential, final decisions about people). This should take one meeting and remove much of the guessing.

2. Make disclosure boring

If admitting AI use feels like a confession, people won’t do it. Make it a routine line instead: “Drafted with AI, reviewed by me” on a doc, or a note in the pull request. When it’s normal metadata, nobody reads it as an accusation, and reviewers know where to look harder.

3. One review standard, whoever or whatever wrote it

The author owns the work, full stop. AI-assisted work doesn’t get a lighter review because it looks polished, and it doesn’t get a harsher one because of how it was made. Review the result against the same bar. This single rule can defuse a lot of the tension between fast and careful people.

4. Pair heavy and light users on real work

Workshops rarely change habits. Working next to someone does. Pair a heavy user with a skeptic on an actual task for a week. The heavy user learns where their output needs checking. The skeptic sees what’s genuinely useful. In Microsoft’s 2026 survey, Frontier Professionals were far more likely than other AI users to say their teams share AI tips, learnings and mistakes (61% vs. 36%). Sharing is the habit worth copying.

5. Don’t punish refusers, and don’t let them set the pace

Some people have good reasons: quality concerns, client rules, or simply that they’re fast without it. Respect that. Judge them on outcomes, not tools. But be clear that the team’s estimates and norms will reflect how most of the work actually gets done now, and that skepticism is welcome in review, not as a veto.

Managers Go First

One more finding is worth taping to your monitor. Microsoft’s 2026 report cites a separate study of 1,800 workers globally in which employees whose managers actively modeled AI use reported a 17-point lift in the value they got from AI and a 30-point lift in trust in agentic AI.

That doesn’t mean managers should use AI for everything. It means they should use it visibly, say where it helped and where it didn’t, and show that disclosure is safe. If the person running the team hides their own use, everyone else will too.

Where This Leaves the Rest of the Team

Uneven adoption is hard to manage partly because it’s hard to see. When AI work happens in private chats and separate tools, a manager only notices the split when it surfaces as a fight in review or a missed estimate.

That’s one reason we’re building VirtualStation around a different assumption: the people, the meetings, the tasks and the AI that helps with them should sit in one shared workspace. AI meeting summaries, action items and AI standups live next to the task board and team chat, so the AI’s contributions show up where the team already works instead of in someone’s private tab. It won’t set your norms for you. It does make it easier to notice who is stuck, where work is piling up and what came out of a meeting, whether it was produced by a person or with AI.

Speed Was Never the Hard Part

Getting individuals to move faster turned out to be the easy part. Keeping a team of people moving at different speeds pointed in the same direction is the real work now.

You will never get everyone using AI the same amount. You can get everyone trusting what the team ships.

If this is the kind of team problem you’re working through, VirtualStation is being built around exactly this idea. You can try it free.

Sources

• Gallup, AI in the workplace indicator, May 2026
• Microsoft, 2026 Work Trend Index, May 2026
• Slack, Workforce Index, November 2024
• BetterUp, AI workslop research, September 2025