The End of the Exponential Is Not the End of Work
The public conversation about AI still behaves as if we are waiting for a single decisive moment. One day, the models will “arrive”, and work will change overnight. Until then, we argue about the curve: how steep it is, whether it is flattening, and who is ahead.
This framing misses the more important point.
Even if we accept that AI capability has been growing at an exponential pace, the end of that exponential – the moment when novelty fades or gains become less dramatic – will not be the end of work. It will be the end of a particular scarcity.
For decades, organisations have treated the production of high-quality knowledge work as expensive. Good analysis, decent drafts, credible plans, clean code, coherent summaries, and structured options have all been constrained by time, attention, and the availability of skilled people. AI changes that. It makes the generation of plausible output dramatically cheaper.
But when output becomes cheap, it does not eliminate work. It relocates value upstream and downstream of production: into judgement, accountability, prioritisation, risk management, and organisational trust.
The technology changes the economics of doing. It does not remove the necessity of deciding.
The capability curve is not the adoption curve
There are at least two curves unfolding simultaneously.
The first is the curve that captures model capability: better performance with more compute, broader data, longer training time, and improved objective functions. This is the curve that dominates headlines and executive briefings. It is also the curve that can feel almost uncanny, because it improves in discontinuous leaps. A system goes from “impressive” to “useful” to “genuinely competent” faster than most organisational planning cycles.
The second curve is the curve of adoption inside institutions. This curve is governed less by the sophistication of the model and more by the sophistication of the organisation: the quality of its data, the clarity of its decision rights, the maturity of its governance, the coherence of its risk posture, and the degree to which incentives align with desired behaviours.
The first curve creates possibility. The second curve determines outcomes.
This is why the future will not be defined simply by who has the best model. It will be defined by who can convert capability into reliable performance inside real operating environments.
Why “more capability” does not automatically mean “less work”
The intuition that AI should reduce work is understandable. If a system can draft, analyse, code, and propose options in seconds, surely the human workload collapses.
In practice, the first-order effect is often the opposite: more activity.
When output becomes abundant, organisations face three predictable pressures.
First, the volume of decisions increases.If teams can generate ten plausible strategies rather than two, leadership must decide between ten. When a tool can produce five product roadmaps, governance must select one, defend it, and live with its consequences. Abundant output expands the option set. It does not choose.
Second, quality assurance becomes more central, not less.When work is produced quickly, it is easier to ship something that is almost right. In regulated or high-stakes environments, “almost right” is often worse than “not shipped.” The work shifts from producing the first draft to verifying, stress-testing, and controlling the second-order impacts.
Third, coordination costs rise unless the operating model changes.If each function adopts AI independently, the organisation generates more artefacts, more interpretations, more dashboards, and more “insights.” Without shared standards and decision rhythms, the result is not productivity. It is noise.
In other words: AI reduces the cost of generating output. It does not reduce the cost of aligning people around a direction, managing risk, or building trust. Those remain distinctly human challenges – and they are the challenges that determine whether output becomes value.
The verified-task illusion
Some domains will move faster than others, and this can distort executive expectations.
Coding, for example, benefits from an external scaffold. Repositories, tests, compilation, and review processes provide immediate feedback. In effect, the work is partially verifiable. It is easier to close the loop.
Many leaders generalise from these domains and expect similar results in the broader economy. But much of enterprise work is not verifiable in clean, binary terms. It is ambiguous by design. What is the correct operating model after a merger? Which trade-offs are acceptable between speed, control, and resilience? How do you balance local autonomy with global standardisation? These are not questions with crisp reward signals.
AI will get better in these areas, but the constraint is no longer only technical performance. It is institutional clarity. Organisations must decide what they value, how they measure it, and who is accountable when trade-offs go wrong.
The danger is that tech-led adoption treats ambiguity as a deployment problem: install the tool, train the users, and expect outcomes to follow. In ambiguous work, that approach fails because it bypasses the core requirement: decision design.
The moment the hype slows, the operating model becomes visible
The public fascination with exponential curves will not last. It never does. The curve may still rise, but the organisational perception of it will flatten as novelty becomes normal. At that point, competitive advantage will be determined less by who has adopted AI and more by how well they have reorganised around it.
This is the overlooked implication of the “end of exponential” narrative.
When the hype settles, weak operating models are exposed. Organisations that treated AI as a tool roll-out will have implemented capability without control: more outputs, more iterations, more speed at the edges, but no material improvement in enterprise throughput. The constraints will simply move: from production to governance, from drafting to decision making, from knowledge to trust.
Organisations that treated AI as an operating model shift will have done something different. They will have redesigned workflows, clarified decision rights, built feedback loops, and invested in adoption as a behavioural change programme. Their advantage will not be that they generate more. It will be that they decide better and execute faster, with fewer costly errors.
Why the future is business-led
If this argument is correct, the next phase of AI will reward business leadership more than technical leadership.
That is not a critique of technology teams. It is a recognition of where value resides.
Technology teams can deploy systems. They can secure them. They can integrate them. But they cannot define value in isolation, because value is contextual: it is customer outcomes, cycle time, risk reduction, margin, resilience, and reputation. Only the business can decide what to optimise, what to trade off, and what failure looks like.
Business-led AI means:
starting with the workflow, not the model
defining decision rights and escalation paths before automating steps
making the safe path the easy path, so teams do not bypass controls
measuring outcomes rather than activity
treating trust as an operational constraint, not a cultural aspiration
This is also why the future must remain human-centred. Not because organisations are sentimental, but because accountability cannot be automated. The moment something goes wrong – a compliance issue, a customer failure, an operational incident – the organisation will not hold a model responsible. It will hold leaders responsible.
Human-centred implementation is the only implementation that scales, because it acknowledges that people remain accountable for decisions even when machines generate the options.
The new scarcity: judgement, not output
The end of exponential novelty will not end work. It will end the scarcity of producing plausible artefacts.
The next scarcity is judgement: the ability to choose, to prioritise, to manage risk, and to align people around decisions that must be made under uncertainty. Organisations that invest in decision clarity and governance will translate AI capability into performance. Those that do not will drown in abundant output and wonder why the promised productivity never arrived.
In the coming years, competitive advantage will not belong to the organisations with the most AI tools. It will belong to the organisations that can combine machine speed with human accountability – and that have the operating model to make that combination reliable.
That is not the end of work.
It is the beginning of a different kind of work, where leadership, not technology, becomes the decisive bottleneck.
Comments