What happens when human attention becomes the limiting factor in an intelligence-abundant organisation?

For years, we've worried about whether we have enough intelligence in our organisations.

Enough expertise.

Enough information.

Enough analysis.

Enough people capable of solving difficult problems.

AI changes that equation.

We are moving towards organisations in which intelligence  or at least access to extraordinary analytical, generative and problem-solving capability may become abundant.

So what becomes scarce?

Increasingly, I think one answer is:

Human attention.

And that has much bigger implications than simply telling people to manage their time better.

We may be getting an early glimpse of an intelligence-abundant organisation

Anthropic published a fascinating piece of research last week looking at what has happened inside its own organisation as AI agents have become deeply embedded in work.

Anthropic is obviously not a typical employer. It is one of the organisations closest to the AI frontier, so we shouldn't assume everybody else will follow the same path.

But that is exactly why I find it interesting.

It may allow us to see some things earlier.

By August, AI agents were working six times as many hours as the humans directing them.

Output increased.

Engineers produced more code. Salespeople closed more sales. Legal and finance teams processed more work.

But then they hit another constraint.

The humans.

Even among non-technical employees, reading all the output produced by their interactive Claude sessions would take more than half their working week.

The AI could produce faster than the human could absorb.

And Anthropic describes human attention becoming the bottleneck.

That creates a very different workplace problem.

We've spent years asking how AI might save human time.

What happens when it can create more work, analysis, ideas, possibilities and decisions requiring our attention than we have time to consider?

This isn't simply overload

I think this distinction matters.

We've always been capable of giving people too much to do.

AI introduces something slightly different.

One person can now initiate multiple streams of intelligent work simultaneously.

Research this.

Analyse that.

Build this.

Compare those.

Challenge this strategy.

Run these scenarios.

While you're getting on with something else, the machines are working.

That's extraordinary leverage.

But all those streams eventually come back.

And somebody has to decide:

What deserves my attention?

What can I skim?

What needs deep thought?

What requires judgement?

What can safely proceed without me?

What looks wrong?

When do I intervene?

What can I ignore altogether?

This is why I've become increasingly interested in Attention Allocation.

If intelligence becomes abundant, the human advantage isn't going to come from trying to consume all of it.

It may come from becoming extraordinarily good at deciding where finite human attention creates the most value.

And the human role begins to change

There was another part of the Anthropic research that I found even more significant.

As AI took on more execution, the capabilities the organisation valued in humans began to change.

Hiring managers increasingly emphasised autonomy and judgement, rather than simply technical domain expertise and implementation.

Employees described their own work moving away from doing everything themselves towards managing AI, making decisions, directing work and dealing with more strategic questions.

That's worth thinking about.

Because the human capabilities becoming scarcer may not be additional technical skills.

They may increasingly be:

Judgement.

Autonomy.

Strategic thinking.

Orchestration.

Interpersonal coordination.

Adaptability across domains.

The human role doesn't necessarily disappear as machine capability increases.

It moves.

From doing all the work towards judging, directing and orchestrating intelligence.

And I think that has enormous implications for how we develop people.

Being able to use AI isn't the same as being capable of directing it

We are currently putting huge amounts of effort into AI skills.

Quite rightly.

People need to understand the tools.

But imagine one person supervising five or ten intelligent agents working simultaneously.

The limiting capability isn't necessarily whether they know which button to press.

Can they frame the right problem?

Can they tell good work from plausible nonsense?

Can they spot what is missing?

Can they decide which output deserves deeper investigation?

Can they hold several competing possibilities without becoming overwhelmed?

Can they challenge the machine?

Can they integrate different perspectives?

Can they make the final judgement?

And can they take responsibility for what happens next?

That's a very different definition of being “good at AI”.

It is one of the reasons I believe Human Frontier Thinking® matters.

The six Human Frontier Capacities: Edge Sensing, Courageous Curiosity®, Independent Thought, Dynamic Deliberation®, Co-Creation and Adaptive Action  weren't designed as a set of AI skills.

But the more I look at the human demands emerging around advanced AI, the more relevant they become.

Notice what is changing.

Question what you know.

Form your own view.

Judge what matters.

Think with other intelligence.

Act, learn and adapt.

Those are increasingly valuable human capabilities precisely because extraordinary machine intelligence is available.

When intelligence becomes abundant, knowing where to allocate it becomes a leadership capability

There is an organisational version of this too.

If AI can do something, should it?

Should the human do it?

Should they do it together?

Who should think?

Who should challenge?

Who decides?

Who has the authority to act?

Who remains accountable?

And what happens to human capability as a consequence?

That's what I mean by Intelligence Allocation.

The question is no longer simply:

What can AI do?

It becomes:

How do we allocate the work and the authority between human and artificial intelligence?

And there are always two tests.

Performance today.

What produces the best result now?

And:

Capability tomorrow.

What ensures that humans continue developing the understanding, judgement, agency and expertise we'll need in future?

Because an organisation can make an apparently brilliant efficiency decision today and quietly create a capability problem for five years' time.

Which brings us to young people

This is the part I think we need to take very seriously.

Anthropic's share of junior technical hires has fallen markedly as its use of AI has increased.

Again, this is one unusual company. It would be irresponsible to turn it into a prediction for the whole labour market.

But it adds to a pattern we've been watching elsewhere.

AI is becoming particularly capable of doing some of the routine work traditionally given to people at the beginning of their careers.

And much of that work was boring.

I certainly wouldn't argue that we should preserve pointless work because previous generations had to endure it.

But some of those tasks were doing something else at the same time.

They were developing the human.

You did the first draft.

You researched the background.

You watched somebody more experienced.

You got something wrong.

You saw hundreds of examples.

Patterns gradually became visible.

Your confidence grew.

Your judgement improved.

Eventually, you became the person who could look at something and know that it didn't quite feel right.

That is how expertise develops.

So:

Who becomes tomorrow's expert if AI performs the work through which today's experts became expert?

This is not an argument against automation.

It's an argument for being much more deliberate about human development.

Don't protect the task. Protect the capability the task was developing.

If AI removes the task, find another way for the human to get the reps.

And what happens to the established expert?

There's another human transition happening at the other end.

A new peer-reviewed study of more than 2,000 Finnish workers found something quite nuanced.

Frequent AI use was associated with greater confidence in people's ability to learn and use AI.

Good.

But AI use, organisational AI deployment and particularly AI-induced identity threat were also associated with greater AI-related strain.

People can therefore become more capable with AI while simultaneously experiencing uncertainty about what AI means for who they are professionally.

I think organisations underestimate this.

Imagine spending 20 or 30 years becoming excellent at something.

Your knowledge matters.

People come to you.

Your expertise is part of your identity.

And suddenly some of what made that expertise valuable can be produced in seconds.

Teaching that person how to prompt isn't necessarily going to answer the question underneath:

If what made me valuable is changing, where does my value come from now?

That's part of what I've been exploring through Endangered Expert Identity®.

Adaptation isn't purely technical.

Sometimes people have to let go of who they were in order to discover what else they could become.

There is an authority problem too

IBM's new UK research gives us another part of the picture.

Eight in ten executives surveyed said an AI-related incident had reached their board or C-suite during the previous year.

And 68% expect AI agents to have some level of decision autonomy by 2027.

Yet only 53% believe their governance structures can scale to manage that future.

We can write governance frameworks.

We can retain human sign-off.

We can declare that a human remains accountable.

But there is a problem if the human holding that authority no longer has sufficient understanding, attention or judgement to exercise it meaningfully.

This is why I've argued:

Human-in-the-loop is not the same as human-judgement-in-the-loop.

And it is why The Human Override® isn't simply about keeping a person somewhere in the process.

The human needs the capability to say:

No. I see this differently. And here's why.

Human authority without human capability isn't meaningful human control.

So what should leaders do?

I don't think the answer is slowing AI down so humans can keep doing everything they did before.

That would miss the opportunity completely.

AI can remove enormous amounts of work that doesn't deserve human time.

Let it.

But leaders now have another responsibility.

They need to become much more deliberate about what happens to the human when the work changes.

Where should scarce human attention go?

Which decisions still require human judgement?

What should people stop doing?

What capabilities must they continue exercising?

How will junior people develop expertise if the old pathway disappears?

How do experienced people reimagine their value?

What authority are we giving AI — not merely what work?

And if humans remain accountable, are we developing them sufficiently to exercise that accountability?

This isn't primarily an AI training agenda.

It's a human capability agenda.

Perhaps we're approaching the really interesting part of the AI story

For the last few years, we've been understandably fascinated by what these machines can do.

I am too.

But I'm becoming increasingly interested in what their capability requires from us.

If machines can produce almost limitless information, perhaps we need better discernment.

If they can execute at extraordinary speed, perhaps we need better judgement.

If they can run multiple streams of work simultaneously, perhaps we need better Attention Allocation.

If they can do more of the work, perhaps we need better Intelligence Allocation.

If expertise becomes easier to access, perhaps we need to become better at asking questions, making connections and seeing what others haven't.

And if AI keeps changing what humans do, perhaps our most important capability is learning how to keep becoming.

That's the heart of Human Frontier Thinking® for me.

Not protecting humans from artificial intelligence.

Developing humans capable of doing extraordinary things because this extraordinary intelligence now exists.

The question I've been asking is:

What could happen to our thinking if we learned to work with this extraordinary intelligence really well?

Anthropic may be giving us an early glimpse.

Humans can become extraordinarily leveraged.

But leverage without increased human capability creates a bottleneck.

So perhaps the leadership question for an intelligence-abundant organisation isn't:

How much more can AI do?

It's:

What is the best use of the human here — and are we developing humans capable of doing it?

For the past two years, Carol has been testing a question: can working with AI make her a better thinker? She co-creates with Chad as a creative thinking and ideation partner, and Claudio as an objective sparring partner. Challenging assumptions and testing consistency and clarity. It is a live, ongoing experiment in HI:AI® and Human Frontier Thinking®.