AI and hybrid leadership: the pitfalls, the potential and the culture that decides it

By now, AI has been rolled out in most organisations. Many have an AI policy that describes what you may use and how. But very few have started the conversation about what AI does to collaboration, working habits and team culture. Here are the pitfalls we see right now, the potential we think is being underused, and the culture that decides how well we build it into everyday work.

The meeting that ended twice

A leader we’ve worked with called us recently. She’d had a difficult conversation with an employee. Not a dismissal, but one of those conversations where you have to say something about performance that neither side is looking forward to. HR joined over Teams, perfectly normal in 2026.

The conversation actually went well. The employee said thank you, got up and left the meeting room. But the leader and HR carried on talking afterwards, drew conclusions from the reaction and discussed what should happen next. Just as people often do in a meeting room once the door has closed. They talked about how the meeting had gone, how the employee had taken the message, and what they’d do next if things didn’t improve. All the things you only say when you think you’re alone in the room.

When the Teams call ended and the leader was back at her desk, the automatic meeting notes landed in the inbox of everyone invited to the meeting. With a summary, conclusions and action points. Including everything that had been said after the employee left the room.

There’s no villain in this story. Nobody pressed the wrong button. Nor was there really a fault in the system. Access to the notes follows the invitation, not attendance, and that’s exactly how the tool is built. The problem was that nobody had talked about it in advance, or agreed a standard for how a follow-up meeting should work in a world where a virtual assistant listens in and takes notes for everyone. In this case the damage was limited. The leader immediately reached out to the employee and explained what had happened, and no sensitive information had been shared. But we can all imagine scenarios that could be far worse.

The silent participant nobody has made a decision on

On the surface, the story is only about meeting notes. Underneath, it’s about the fact that one more participant has joined our meetings, and nobody has decided what to do about it. It says nothing along the way. It doesn’t raise its hand to point out that the conversation is becoming confidential. It listens and writes things down, even when we’ve long since forgotten it’s there. We’d never let a person sit in like that without first agreeing what their role was.

That’s the step we think we’re skipping. We roll out tools that reach deep into the way we talk to each other, and then settle for a policy on data and legal matters. But what decides whether it goes well is something else: how do we behave when the assistant is on, what’s the standard for a follow-up meeting, and is it even all right to have someone sitting in who listens without making a sound? Those aren’t questions IT can answer. They’re something the team has to agree on.

That’s the point we want to carry all the way through. If you have clear agreements, a clear core task and a culture where people dare to say things out loud, AI will pay off sooner than you think. If the direction is unclear, meeting discipline thin and trust fragile, AI becomes a magnifying glass on exactly that. In a hybrid team, the road from magnified to broken is just shorter, because we can’t read each other across a desk afterwards.

Microsoft’s 2024 Work Trend Index put figures on how far ahead employees were: three out of four knowledge workers were already using AI at work (and surely that’s only grown since?), and almost as many, 78 percent, said they bring their own tools to work because they’re not willing to wait for the organisation to catch up. The figure we keep coming back to is a third one: 52 percent are reluctant to say they’ve used AI for their most important tasks.

So half your people are probably not quite telling you how the work actually gets done. Not because they’re cheating, but because they’re afraid of looking less capable. That’s a textbook sign of low psychological safety, and that’s where we have to start.

Five pitfalls we’re seeing right now

1. The notes that share more than anyone asked for

The story above isn’t unique. Automatic transcription and summaries are switched on by default in many places, and very few teams have decided who has access, how long the notes live, and what happens to what’s said after the official part of the meeting. Have that conversation before you need it. Agree, too, when recording gets switched off: staff reviews, conversations about a colleague and anything to do with health don’t belong in a searchable text file. We also think there’s a bigger IT conversation to be had about how these virtual note-takers can be made more visible to participants.

2. Faster answers to a bad process

This is the one that costs most, and the one fewest people talk about. We give employees a copilot, and the inbox gets easier to manage and email quicker to answer. But the heavy, clumsy process behind the email is completely unchanged. In other words, we’ve put a turbocharger on poorly thought-through processes and made the pile of email easier to handle, instead of first cutting the number of emails we send.

The UK civil service trialled Microsoft 365 Copilot with 20,000 staff. On average, participants reported saving 26 minutes a day, which is respectable. But 17 percent saw no saving at all. In the Australian government pilot, only about a third used the tool daily. It isn’t the technology that’s failing there. It’s that nobody asked which processes actually needed to change.

On top of that comes what researchers from Stanford and BetterUp have named workslop: AI-produced material that looks polished but lacks substance, so the recipient has to redo the work. 40 percent of those surveyed had experienced it, and each incident took just under two hours on average to clean up. No time has been saved. The time has just moved to a colleague.

3. Shadow AI and a digital A team and B team

When 78 percent bring their own tools, two things happen. One is what the IT department worries about: customer data and confidential agreements ending up in services nobody has approved. The other is more of a leadership problem. The tricks each person figures out at home become private property. In the past, you discovered a colleague’s clever way of doing things by looking over their shoulder at the office. If people sit on their own, quietly prompting an AI that works better than the one the company provides, the head start may become personal rather than shared, and six months later you have an A team and a B team that management struggles to explain from a distance.

4. Distrust of what we’re sent

Distrust is building up, and it’s about more than AI. It’s about whether we can rely on everything we’re sent. Have the figures in the memo been checked, or do they just look plausible? Has a person thought the analysis through, or was it generated in thirty seconds? We’ve started reading each other’s work with an extra filter on, and we don’t say so out loud.

“Did my colleague even write this themselves?”

That suspicion isn’t paranoia, it’s been measured. In the workslop study, 42 percent said they saw a colleague as less trustworthy after receiving AI-generated work without substance, and 54 percent as less creative. That’s where trust leaks, and in a hybrid team it leaks silently, because nobody has the conversation by the coffee machine afterwards.

The danger isn’t that the suspicion arises. The danger is that it stays. Nobody writes in a Teams thread that they’re not sure whether a colleague actually thought this through. You just quietly turn down your trust in the next piece of work. That’s why it has to come out into the open, and that’s a leadership job. Say what you use AI for, say just as clearly what you stand behind yourself, and make it something people can ask about without offending anyone.

5. The new interruption: ten threads, all waiting for you

We’ve felt this one ourselves. It’s become easy to have ten chats running in parallel, in several tools, across tasks that have nothing to do with each other. Each thread feels productive. Together they’re a context-switching machine, and it can get properly stressful to conduct an orchestra you hired yourself.

It’s the same mechanism we already know from hybrid work: too many channels, too many interruptions, and the worst are the ones we cause ourselves. We’ve written about it in Meeting overload and in our piece on the digital work environment. AI can easily make both problems worse unless we decide how to use it. An assistant that can suggest, summarise and remind us of things is also a new source of interruptions. Decide when it’s on.

The potential, if you tackle the process instead of the tool

By now we probably sound like we’d rather do without it. We wouldn’t. We think AI is the biggest opportunity knowledge work has had in many years, and that the work can become markedly more interesting once the noise goes. But the gain isn’t in the licence. It’s in what you choose to use it for.

Choose by business value, not by excitement about the technology

The AI initiatives we’ve seen work all have one thing in common. They started with a question to employees: which of our processes are heavy, pointless and soul-destroying, and what would you get rid of first? It isn’t just good involvement, it’s the only way to find the right places to act. Management only knows the tip of the iceberg, and there’s good evidence for that. Then prioritise by what creates value for the customer or the core task, not by what gives the most impressive demo or sounds most technically advanced.

Meeting preparation that actually happens

We’ve all sat in the meeting where the first ten minutes go on bringing everyone up to date, because nobody had time to prepare. It’s one of the places where we get most out of AI ourselves: a summary of what’s happened since last time, what’s still outstanding, and what we should have a view on before we walk into the meeting.

With one important caveat. It still has to be our preparation, supported by a machine. Not the machine’s opinion that we nod along to. If everyone turns up with the same AI-generated summary and no view of their own, we haven’t improved the meeting, we’ve just automated agreement. You’ll find more ways to handle this in meetings and meeting culture.

Knowledge sharing we’ve talked about for ten years

Knowledge sharing is the classic good intention that everyone agrees with and nobody gets round to. For the first time, there are tools that can make it cheap: notes from knowledge-sharing sessions that actually get written, and a shared model that can gather it up, search across it and answer when someone asks six months later.

It isn’t just a good idea, there’s evidence behind it. In the best-documented study in the field, 5,172 customer service agents followed over time, the number of cases resolved per hour rose by 15 percent. The interesting part is why: in practice, the assistant captured the tacit knowledge of the experienced agents and passed it on to the less experienced ones. New people got good faster. That’s a big part of the point of hybrid resilience: knowledge mustn’t sit with individuals alone.

The shared AI beats the staff handbook

Where do you book catering for a meeting? How do we usually approach a quote? Where’s that template? Those are exactly the questions a new employee doesn’t dare ask a third time, and that hybrid working makes harder to get answered, because there’s nobody sitting next to you. A shared AI that everyone is responsible for keeping up to date is a far better answer than a handbook that was last updated in 2023.

But be honest about the downside. The out-of-date handbook nobody reads may be harmless. An out-of-date answer, delivered by an AI with complete confidence, isn’t. And because the assistant respects the access rights you already have, everything that’s been shared too widely by mistake suddenly becomes much easier to find. Clean up permissions and old content before you switch it on. It’s dull work, and it’s what decides whether it works.

How we use it ourselves

We trust advice most from people who’ve got their own hands dirty, so here’s our own practice. We work with the concept of hybrid intelligence, and the principle is simple: we take the lead, the machine takes the boring part.

• Think first, then write a brief. We sit down and work out what we think about the subject before we open anything at all. Then we write it down rough and unstructured, without worrying about spelling or order, and use AI to structure it, polish it and make the presentation tidier. The view is ours. The machine helps with the form.
• A daily summary. Which important emails have landed, what’s outstanding, what’s in the calendar. Not because we can’t read our own inbox, but because we’d rather use the sharpest hour of the morning on something better than trawling through 100 new unread messages.
• Research and monitoring as automation rather than chat. Keeping an eye on what’s happening with customers and in the market shouldn’t depend on remembering to ask. Set it up so it arrives by itself.
• Orchestration, when AI handles a whole task. The moment it doesn’t just help but delivers something, it should be treated as a contribution from a colleague. Who follows up on it, when, and how does it get into your project management? Without that link, you end up with a pile of output nobody owns.

And a warning to ourselves, passed on free of charge: the better we get at this, the more threads we have running. The discipline isn’t using AI. The discipline is finding a solid structure for not working on ten different tasks at once.

The culture that decides it: the agreement, not the policy

Most organisations we visit have an AI policy. Very few teams have an AI agreement. The difference is crucial. The policy protects the organisation and is about data, legal matters and risk. The agreement protects collaboration and is about what we expect from each other. It’s the same distinction as between a working-from-home policy and a hybrid team charter, and on an ordinary Tuesday, it’s the team’s own agreement that does the work.

Set aside an hour at your next team meeting and get specific about:

1. What do we use AI for, and what do we do ourselves? Which tasks won’t we automate, because they’re the craft itself, or because the quality of our thinking depends on them?
2. When do we say so? Should it be clear when a paper was made with help? And to whom? Just agreeing on it removes a lot of the suspicion.
3. Who is responsible for the content? The answer is always the sender. It just needs saying before it matters.
4. What must never go into a prompt? Personal data, confidential agreements, colleagues’ health.
5. What do we do with meetings that are recorded? When is recording switched off, who has access to the notes, and how long do they live?
6. How do we share what we find out? Where do we put the tricks and prompts that work, so the gain belongs to the team and not just to one person?
7. Where should the time we save go? Decide in advance. Otherwise it gets eaten up by more meetings, more messages and faster response times, and you’ve bought a subscription to being busy.

Notice that none of the seven questions are technical. They’re all about trust, direction and what we expect from each other. And they get considerably harder to raise if you wait until something unfortunate has happened.

Start with the conversation, not the tool

We’re not going back to the time before assistants. There’s too much to gain: work that gets far more exciting once the noise is out of the way, and a business that gets sharper at the same time. But we shouldn’t repeat the mistake we just made. When hybrid work arrived, we rolled out the technology first and talked about collaboration afterwards. It cost us a couple of years of organisational corrosion.

So start small. Take the seven questions to your next team meeting. Not as a policy, but as a conversation. We’ve yet to see a team where that hour didn’t uncover something worth knowing. And if you’d like a hand with the process, that’s exactly what we work on in team development, as a workshop or a talk. Write to us here, and we’ll get back to you with a concrete proposal.

Sources

• Microsoft and LinkedIn (2024), Work Trend Index: AI at Work Is Here. Now Comes the Hard Part.
• Niederhoffer et al. (2025), AI-Generated “Workslop” Is Destroying Productivity, Harvard Business Review, with BetterUp Labs and Stanford Social Media Lab.
• Brynjolfsson, Li and Raymond (2025), Generative AI at Work, Quarterly Journal of Economics.
• Dell’Acqua et al. (2026), Navigating the Jagged Technological Frontier, Organization Science.
• UK Government Digital Service (2025), Microsoft 365 Copilot Experiment: Cross-Government Findings Report.
• Australian Digital Transformation Agency (2025), Evaluation of the whole-of-government Microsoft 365 Copilot trial.