What AI Is Teaching Us About Organisations.

© Rich Baker

If you've worked in organisations for long enough, you've probably been in that meeting.

Someone shares a better way of doing something – not a vague "wouldn't it be nice" idea, but a practical improvement. Something that saves time, improves quality, or simply makes work easier.

For a moment, it feels like exactly the kind of discovery every organisation says it wants. People ask questions. Someone spots another application. Someone offers to try it. The discovery grows because others build on it. Before long, it no longer belongs to one person.

But sometimes something very different happens. The questions become polite rather than curious. The discovery stays exactly where it started: with the person who made it.

I've seen both versions of that moment more than once. The more I've thought about them, the less I think they're meeting problems. They're organisational ones. Because the issue isn't that someone discovered a better way of working. It's that the organisation didn't.

That may be one of the biggest challenges organisations face as AI becomes part of everyday work.

From "will people use it" to "what happens to what they learn"

For the last couple of years, we've talked a lot about AI adoption. Would people use it? Trust it? Would it replace jobs, or turn out to be another overhyped technology?

By mid-2026, AI is no longer a novelty. It's part of everyday work – writing, summarising, analysing, planning, coding, searching. Microsoft's research into anonymised Copilot conversations points to these as among its most common uses in knowledge work. Adoption across companies is still uneven, but no longer marginal: recent U.S. Census Bureau survey data has put business AI use at around one in five firms, with much higher uptake in professional services and tech.

So the question is changing. It's no longer simply "will people use AI?" – many already are. The better question is: what happens to what they learn while using it?

That's where I think the real story sits.

Thousands of small experiments, most of which go nowhere

Across organisations, employees are running their own small AI experiments every day. Someone's found a better way to prepare for a meeting. Someone's worked out how to summarise a long document without losing the nuance. Someone's learned that one tool is good for drafting, another for thinking, another for recall. Someone's discovered a workflow that saves an hour a week, or a day.

Most of these discoveries don't look like innovation in the way organisations usually recognise it. They're not big launches or transformation programmes – they're small, situated improvements in how work gets done.

But together, they're potentially enormous. The problem is that most of them never become organisational knowledge. They stay with the individual – in someone's head, browser history, private prompts, personal way of working. They might get mentioned to a close colleague or dropped into a call in passing, but they rarely become part of how the organisation learns.

That isn't really an AI problem. It's an organisational one.

Tacit knowledge, moving faster than we've ever had to absorb it

For decades, "knowledge management" has mostly meant storing information: policies, templates, playbooks, FAQs. Useful, but static – a library of what the organisation already knows.

What employees are discovering with AI is different. It's tacit knowledge: the practical, hard-to-explain understanding that comes from doing the work. The judgement that says "this tool is good for that, but not this." The instinct for how to frame a better question next time.

We've always had tacit knowledge in organisations. It's why experienced people are valuable, and why replacing someone is rarely as simple as replacing a role. When people leave, they take judgement, context, relationships, shortcuts – the invisible value they created, which the job description never captured.

Most organisations only realise what someone really brought once they've gone.

It has always been there, quietly propping up how organisations actually function. AI doesn't create it. It simply makes it move faster, and makes its absence more obvious when it isn't shared.

What's different now is speed. AI is increasing the velocity of individual discovery – and I'm not convinced organisations are evolving at the same pace.

That creates a new kind of gap. Not simply between organisations that use AI and those that don't, but between how fast individuals are learning and how fast organisations can absorb it.

Objectives aren't a learning system

Employees are experimenting, adapting, improving their own workflows. Meanwhile the organisation is often still trying to capture learning through annual objectives, training completion rates, usage dashboards and the occasional success story.

Those things have a place. But they can't capture the dozens of small discoveries people make throughout the year – the failed experiments, the useful shortcuts, the quiet improvements that gradually change how work gets done. Objectives are an accountability mechanism, not a learning one. Many organisations still treat them as the same thing.

If every employee is independently working out how to use AI, every organisation is effectively funding thousands of small R&D projects. The invisible value is already being created. The real question is whether anyone is collecting it.

Why culture is the sticking point

Culture isn't what an organisation says it values. It's what it rewards, protects, tolerates and repeats.

Most organisations say they value collaboration and knowledge sharing. Yet performance systems often reward individual delivery, visibility and measurable contribution. That creates a tension: if I discover an AI workflow that makes me significantly faster, why would I give it away? If part of my value comes from being the person who does something better or quicker than others, sharing it may cost me my edge even if it helps the organisation. That's not selfishness – it's rational behaviour inside a system that rewards individual performance more clearly than shared capability.

AI makes that contradiction harder to ignore, because the value of sharing may now dwarf the value of individual improvement. One employee saving two hours a week is useful. That same employee sharing the workflow so 500 colleagues each save twenty minutes is far more valuable to the organisation – yet most performance systems still recognise the first far more clearly than the second.

In an AI-enabled workplace, the highest-value employee may not be the one who produces the best work in isolation, but the one who makes everyone else better. That calls for a different question in performance conversations: not just "what did you deliver?" but "what did you help the organisation learn?" Harder to measure – but organisations have never improved by measuring only what's easy to count.

What this could look like in practice

If organisations want AI to build real capability, learning needs to travel continuously, practically and socially – not once a year in a case study. That might mean team rituals for sharing useful AI workflows, communities of practice focused on redesigned work rather than clever prompts, recognition for people who scale their discoveries through others, and leaders asking not just "are you using AI?" but "what have you learned that the rest of us should steal?"

It also means being honest about where we actually are. For most people, AI still means typing into a box – a summary, a draft, an idea check – not some fully agentic future doing complex work unsupervised. Useful, sometimes powerful, but still heavily dependent on the human using it. Which is exactly why the human learning matters so much.

We've seen this pattern before. Search engines were available to everyone, but some people got much better at finding what mattered. Spreadsheets were available to everyone, but some used them to transform decision-making. The technology becomes ordinary; the capability doesn't. That's likely what's happening with AI – access becomes less distinctive, and advantage moves to whoever learns fastest around it.

Internal comms has to change shape

This is also where internal communication has a role. For years it's largely meant cascading information: leaders decide, communicators explain, employees receive. AI adoption needs something more networked – the job isn't just telling people what the organisation wants them to know, it's helping the organisation hear what its people are learning.

That's a different posture. It means comms becomes part of the learning system itself – surfacing useful practice, connecting people, making new behaviours visible. Adoption is social: people change faster when they see someone like them using a tool in a way that clearly helps. A practical workflow from a trusted colleague will shift behaviour faster than another broad message about the importance of AI.

Training gives people a starting point. Governance gives them boundaries. Tools give them capability. Culture determines whether learning compounds.

The real limiting factor

Two organisations can buy the same technology, roll out the same licences, run the same training, and use the same language about transformation. A year later, one may be only marginally more efficient, while the other has fundamentally changed how work gets done. The difference won't be the software. It'll be the learning system around it.

Once individual learning starts outpacing organisational learning, the technology is no longer the limiting factor. The organisation is. Leadership, culture, communication, incentives and performance management all determine whether individual discoveries spread or stay put.

Redesign, not just disappearance

This is also why the job-loss debate, while important, doesn't quite capture what's happening inside organisations. The evidence on AI's labour market impact remains mixed and concentrated in particular sectors and use cases – Goldman Sachs economists have described the effect so far as "visible but narrow" rather than broad-based.

For many workers, the more immediate change isn't disappearance – it's redesign. Tasks shift. Standards rise. First drafts get cheaper. Average work gets easier to produce. Being able to use AI becomes less impressive; being able to judge, adapt, combine, challenge and share what works becomes more important.

Culture is part of that. So is leadership, performance management, communication and incentives. AI hasn't made those things more important – it's made them impossible to separate.

The questions worth asking

AI is revealing the operating system of the organisation itself.

Not just culture, but leadership, performance management, communication, incentives, governance and learning – the whole set of mechanisms that decide whether people feel safe to experiment, whether they're willing to share, whether performance systems reward generosity or hoarding, and whether scattered individual progress ever becomes collective advantage.

We spent the first phase of AI at work asking whether people would adopt the tools. The more important question now is whether organisations can adopt the learning.

The competitive advantage was never sitting inside the technology. It's sitting inside thousands of people quietly reinventing how they work every day, creating invisible value the organisation hasn't yet learned to see.

Whether invisible value becomes organisational capability isn't a technology question anymore. Ultimately, it's a leadership one.