Intelligence is Becoming a Commodity; Imagination Will Be the Advantage

As I worked towards my PhD in AI,  I watched the field change from both sides. Early AI was an expensive endeavour: you needed data, you needed to train the model, and models drift — the one that gives you value during training is not necessarily the one that serves you in production. Machine learning asked humans to find the features that mattered. Deep learning let the networks discover the latent relationships themselves, but demanded enormous data. Generative AI threw that bottleneck away, and industry began implementing AI everywhere.

What I see ahead is the number of interceptions AI can make only growing. AI is becoming the new electricity of our generation. Every digital and physical infrastructure that surrounds you will have AI in it — a technological revolution humanity has not seen before.

At every inflection point, enterprises have had to accept a new way of working that the technology demanded. With AI, the magnitude of that change is going to be far greater than we have ever anticipated.

The change will not stop at how we run our businesses. AI will optimise processes. It will improve decision-making by helping leaders find a path through an overwhelming number of options. But the deeper shift is in the value a business provides to its customers — that is going to change too.

The whole point of generative AI is that it has the capability to create. So it does not only make an enterprise more efficient; it makes an enterprise more inventive. We are already seeing this in drug discovery, where AI solved the protein-folding problem that had stood for a hundred years. Challenges that were impossible to solve before become solvable.

So I would separate the impact into three layers: optimisation of processes, augmentation of decision-making, and the creation of genuinely new value. Most organisations are working on the first. The competitive advantage will come from the third.

This is the most important question in enterprise AI today. Almost every organisation is using AI and building pilots — I would say around 90% — and only about 6% of those pilots actually reach production.

There are several reasons for that. The technology itself is still maturing. Generative AI has been genuinely accepted for less than three years, and it is being asked to work with systems, data and processes that have been accumulating for thirty. AI has to go and imbibe all of those intricacies inside the organisation before it can run in production. That is not an easy job.

Then there is the difference in the question being asked. When you build a pilot, the question is: can the technology solve this problem? Today the answer is almost always yes — we can solve anything and everything. When you deploy to production, the question changes completely: can this technology follow the policies, the processes and the guardrails the enterprise has put in place, and still function the way we want it to? Sometimes the honest answer is no.

Production demands reliability, adaptability, robustness and trustworthiness. The capability and the accuracy of AI have increased many times over; reliability, robustness and trustworthiness are what we still have to look upon. That is why governance becomes the deciding factor — and it is not a question people ask while they are building the pilot. It should be.

Governance, reliability, robustness and provability have to be in the architecture from day one, and you have to question the design from first principles.

What often happens in the age of AI is that everything moves so fast that people push a solution into the market first and only then start revisiting it. Depending on the industry and the market you are playing in, that is a costly order to work in. If you are in healthcare, life sciences, pharmaceuticals, medical devices, finance, defence or government, you are working inside regulations and norms that are not negotiable. If you have not architected your platform and designed your solution to incorporate that governance, you will have multiple successful pilots and none of them will ever see the light of day.

So my advice to technology leaders is simple: know your industry and know your target market from the first day you start building an enterprise solution. Guardrails and governance are not a compliance layer you apply at the end — they have to be part of the design. Build that way and reliability, scale and measurable value follow. Skip it and you are building demonstrations.

We now know that AI can do the work. Wherever people do their job by interfacing with a computer — the digital workforce — AI can do that work. I know that sounds frightening to many people, but let us not make the mistake of seeing only one side of it. AI is also creating markets that do not exist today, and entire industries we have never had to build before.

On the productivity side, the evidence is already in. Organisations that have deployed AI in production are seeing tremendous gains, especially among developers and engineers. The prevailing wisdom is to use AI as much as you possibly can, and things we thought would take a hundred years are becoming possible in a hundred days.

Operational efficiency is the other half, and there I would say AI is still maturing. Running business processes with AI requires proper guardrails and governance around what can be run by AI and what is too risky to hand over today. Of course the technology will mature. But every organisation has to ask a question first: which processes can be AI-driven, and which processes need AI to augment a human who stays in the loop? Once that line is drawn, the job becomes far simpler and the impact on both operational efficiency and productivity can be impeccable.

This follows directly from the pilot-to-production question, and my answer is the same: the move from experimentation to application is not a capability problem, it is a governance and integration problem.

Experimentation asks whether the model can do the task. Enterprise application asks whether an agent can operate inside thirty years of accumulated systems, data and process, follow the organisation’s policies, and behave predictably every single time. Agentic AI raises the stakes further, because agents act. They can parallelise themselves, they move faster than the processes around them, and they can create backlogs that no human workflow was designed to absorb.

So practical adoption will belong to the organisations that decide clearly where agents can run autonomously and where a human stays in the loop, and that measure their agents continuously — robustness, adaptability, reliability, accuracy, trustworthiness, safety and fairness. The technology is ready to be useful. The enterprise around it is what has to be made ready.

I do not want to jinx it, but organisations carrying less legacy can often bear the fruit of AI more quickly. That makes this a unique opportunity for small and medium-sized businesses. They have less data legacy and less complicated processes, so AI can be implemented and can run far faster than in a large organisation where there is a standard process for everything, where data sits across many sources, and where years of accumulation have made all of it complicated and messy.

So when we talk about strong processes, I would put it slightly differently. What helps AI is not process weight, it is process simplicity. Simplicity in the process and structure in the data are what genuinely empower AI.

That is where technical architecture and data quality decide the outcome. The organisations that looked at this from day one — that kept a simplified structure to their data model and invested in a proper semantic layer — are the ones that will get tremendous benefit from agentic AI. Everyone else will spend their first year of AI work doing data and process archaeology.

I think AI is going to run the business.

We already know AI can do the work. It interacts with computers better than we do, so we are better off leaving that computation to AI. But the more interesting part is value creation. The value a business provides to its customers comes through innovation, and AI will have a significant impact on driving those innovations as well.

So the next generation of enterprise operations, and the next generation of enterprises themselves, is going to be AI-driven. And here is what follows from that. As intelligence becomes commoditised, the competitive advantage of an organisation will no longer come from having intelligence. It will come from how well the organisation uses it. The power of imagination becomes the differentiator.

The first thing organisations have to accept is that their processes need to be redesigned to hold AI and humans in the same workflow. Today’s processes were designed only to keep humans in check. They do not account for AI agents, which can do the job faster and better than a human, which can parallelise themselves, and which can create huge backlogs the surrounding process was never built to absorb. How will these processes work when agents are inside them? That question has to be asked from first principles.

This will not happen on day one. Enterprises are going to take their own time to figure out the processes they need, and that is fine.

But a few parameters have to be in place regardless. Whenever AI is driving a process, you must be able to measure robustness, adaptability, reliability, accuracy, trustworthiness, safety and fairness. Measure them continuously, so that the moment something falls outside the company’s policy it is identified. And the company’s policies themselves have to be reinforced in the system, so that AI is always working within the governance and inside the guardrails. Responsible adoption is not a statement of intent — it is a measurement discipline.

I believe AI is going to drive the next generation of business. Commerce will be completely AI-powered. Innovation will be primarily AI-driven, with humans giving the direction — pointing, guiding, deciding what is worth building.

But every organisation’s intelligence is going to rely on its data model: the centralised model it was able to create, one that captures the behaviours and characteristics that have given that enterprise its competitive advantage over the years. AI will use that and create new value for the customer. Organisations that never built that foundation will find they have nothing distinctive to give their AI.

The way I put it is this: the future of AI is not only AI driving the car. It is AI designing the car itself, building it, and then driving it to take you from one destination to another.