Two billion years ago, one microbe swallowed another. It should have been an ordinary predatory act with one cell the winner and the other erased. Instead, the engulfed cell was put to work, kept alive inside its host, and became the mitochondrion that powers every complex cell alive today.

The “loser” was not eliminated but incorporated into something bigger. It is our oldest lesson that absorbing a rival capability need not be zero-sum.

I mention this because I am concerned that business leaders are constrained by too much zero-sum thinking about AI.

This reductionism shows up as numerous false binaries: agents vs SaaS, America vs China, innovation vs regulation, but most prominently, AI vs human labour.

I will focus on this last topic, which requires exploring three effects of AI on people: substitution, augmentation, and co-creation.

Substitution

All technological revolutions have had a substitutionary effect, and AI will undoubtedly directly replace many human tasks. The received wisdom is that this will continue until entire job roles, and ultimately entire professions, are displaced.

However, macroeconomic data shows that, so far, AI job displacement has failed to materialise at scale, with U.S. executives expecting AI to reduce aggregate headcount by less than 0.4% this year.

The exception is entry-level jobs. Employment of people aged 22 to 25 in the most AI-exposed occupations now trails that of peers in less susceptible jobs by 19%, mainly due to reduced hiring.

The disparity between the prevailing substitution narrative and labour data has made me an avid parser of layoff announcements. AI is the most-cited reason for announced U.S. job cuts this year, yet Gartner judges that most reductions reflect economic conditions rather than automation alone. Press releases typically present cuts as anticipating future AI returns, or mention AI as just one unquantified factor.

Almost no company cites a simple cause-and-effect: AI replaced jobs, so we now need fewer people.

More substitution is no doubt coming, but it is also plausible that something else is going on.

 

Augmentation

AI is not only substituting for humans – it is also augmenting them.

This second synergistic effect is difficult to measure in aggregate, but it stands to reason that if employment has held steady through years of heavy AI investment, then whatever return it is producing is most likely coming primarily through augmentation, rather than substitution.

The early data supports this hypothesis. U.S. labour productivity has grown 2.4% a year since early 2024, against 1.6% before the pandemic, led by the sectors with the most scope for AI – a pattern repeated among Europe’s heaviest AI adopters but not its lightest. Meanwhile, hours worked have barely moved.

In an influential study of AI augmentation in customer service, less experienced staff were “levelled up” – their output rising by as much as a third, narrowing the gap with their most experienced colleagues.

Similarly, the first completed randomised controlled trial of AI in breast cancer screening found that AI-supported reading detected 29% more cancers than standard double reading, with no rise in false positives, while cutting radiologists’ reading workload by 44%. Its final results, published this year, found no increase in cancers missed between screenings.

Ironically, radiology was a prominent field singled out for substitution long before generative AI.

As far back as 2016, the AI pioneer Geoffrey Hinton urged the medical profession to stop training radiologists, predicting deep learning would outperform them within five years.

A decade on, the number of radiologists has grown by 55% at the Mayo Clinic alone, yet demand has outpaced supply. The UK now has a 32% shortfall of consultant radiologists. Hinton himself now expects AI and radiologists to do most image interpretation together.

Beyond medicine, only a fifth of customer service leaders have actually cut staff because of AI; most have held headcount steady while serving more customers. And of the leaders who made redundancies when deploying AI, 55% admit they got those decisions wrong.

Could it be that, while aiming for substitution, we are actually achieving augmentation? And if so, is this a bad thing?

Augmentation could prove a more sustainable source of productivity gains for both companies and the economy – not least because augmentation retains human workers as consumers and taxpayers, avoiding dramatic and costly social displacement.

If we are aiming for substitution but getting augmentation, one obvious shift is to actively target synergy rather than replacement at the outset. For this, we need to get better at measuring it.

But there is a whack-a-mole dynamic at work here. Customer service, where AI has levelled up novices, is also among the occupations where employment of young workers is falling. As the radiology example shows, augmentation requires an increase in demand; otherwise, it turns into substitution by closing the front door.

 

Co-creation

History tells us the greatest economic and societal gains of previous technology revolutions came not by substituting or augmenting humans in existing tasks, but by moving the entire perimeter of possibility.

Most Americans now work in types of jobs that did not exist in 1940, and that new work has followed technologies that complement people, not those that automate them.

Could the real prize of AI be the co-creation of new roles, products, services, companies, industries, and societal categories rather than merely altering or eliminating those we already have? This is the mitochondrial principle of additive integration.

Consider the often-cited case of the railways – but from a substitution, augmentation, and co-creation perspective.

Their first effect was substitution: the canals and stagecoaches, and the trades that served them, were painfully displaced.

Their second was augmentation: existing merchants, manufacturers, and farmers could now reach national markets via the new transport network.

But the real prize was co-creation: entirely new occupations like guards and stationmasters emerged; steel and telegraph firms sprang up alongside the tracks; and, ultimately, novel societal categories like tourism and suburban commuters took hold.

This phenomenon is just becoming visible with AI. LinkedIn counts 1.3 million new roles created since 2023 – AI developers, data annotators, forward-deployed engineers – alongside more than 600,000 data-centre jobs.

DeepMind’s AlphaFold has collapsed protein-structure prediction from years to minutes, mapping virtually all 200 million proteins known to science.

This AI innovation did not merely augment existing scientific methods; it moved the perimeter, co-creating an entirely new industry in AI-native drug design.

Isomorphic Labs, DeepMind’s spin-out, raised $2.1 billion in May, while the first drug whose target and molecule were both found by generative AI entered late-stage trials in July.

AI co-creation is nascent but real.

 

What should leaders do?

Perhaps, then, business leaders are too focused on what can be taken away and not enough on what can be added?

I am not suggesting there is no legitimate or inevitable substitution. There always is. Rather, I am questioning where leadership attention should focus.

Some tangible suggestions:

  1. Examine your AI investments through the lens of substitution, augmentation, and co-creation. What did you set out to do? What actually happened? How much can be measured? Was there another way to look at these projects in retrospect?
  2. What constraints prevent your business from achieving stronger growth and profitability? Could AI augment or co-create a new way to overcome them?
  3. If you were unable to reduce headcount, but could redeploy it, what initiatives would you focus AI on in that world?

Consider putting your most creative strategic thinker on these questions for a month.

 

Conclusion

Anticipating new jobs, professions, and industries is harder than spotting current candidates for automation, but history suggests the former are the real source of future AI returns.

Even if we can’t see the future clearly, we should escape the zero-sum of substitution and augment or create something as well.

The secret to AI productivity may have its roots in two-billion-year-old biology.

 

If you'd like to listen to an audio version of this article, please click here.