Over the past two years, many boardroom conversations about artificial intelligence have followed the same script, according to Dr Ramesh Shanmuganathan, Executive Vice President/Group CIO at John Keells Holdings PLC, Director/CEO at John Keells IT. Someone asks how AI can improve customer service. Someone else wants to know what it will do to headcount. The technology team talks about model selection and governance frameworks. These are useful conversations, he says, but none of them is the conversation that actually matters.
For more than three decades now, enterprise technology has followed a fairly predictable pattern. Paper moved to digital, client-server moved to cloud, disconnected applications moved to integrated platforms. Each shift made the organisation faster and better connected than it was before. But underneath all of it, something stayed constant: new technology was used to run the same organisation a little better. Organisations automated the process that already existed rather than asking whether it should exist at all. They digitised the form instead of asking why the form was needed in the first place.
Artificial intelligence breaks that pattern, but Shanmuganathan believes most organisations have not fully reckoned with how much.
Unlike the technology waves that came before it, AI can reason through a problem, learn from what happens next, and increasingly act on its own judgement without someone standing over it approving every step, he explains. That is a genuinely different kind of capability, and it changes the question that should be asked. Not “how do we bolt AI onto what we have,” but “what would this organisation look like if we designed it from scratch today, knowing what AI can now do?” That second question is uncomfortable. It should be. It’s also the one that will separate the companies that lead the next decade from the ones spending it playing catch-up.
In a conversation with Echelon, Shanmuganathan says the question worth asking isn’t how AI can improve the organisation as it exists today. It’s what the organisation would look like if it were designed today, knowing what AI can now do.
What is the difference between an AI-enabled and AI-native organisation?
Cloud computing didn’t just move servers offsite. It changed how companies thought about capital, risk, and speed. The internet didn’t just connect computers. It rewired how people communicate and where authority sits. I think AI is going to do something similar to the enterprise itself, and organisations that treat it as a feature to switch on are going to be surprised by how much they missed.
Call it the difference between an AI-enabled organisation and an AI-native one. Most companies today are the former. They run pilots, they deploy a chatbot, they roll out a copilot for the sales team, and they call it transformation. It isn’t, quite. Adding AI to an operating model that was built for a slower, more hierarchical world is a bit like bolting a jet engine onto a horse cart. You will go faster for a while. You have not actually changed what the cart was built to do.
There’s a historical parallel here that I find useful, even if it’s a slightly worn one by now. When factories first electrified, most owners simply swapped the steam engine for an electric motor and kept the same floor plan, the same layout, the same production line running in the same shape it always had. It took the better part of a generation before anyone realised that electricity didn’t just replace steam. It let you lay out a factory completely differently, because power no longer had to run through a single central shaft. The productivity gains, when they finally came, came from redesigning the factory around what electricity made possible, not from swapping the engine.
AI is our electricity moment, and I suspect we are still mostly at the stage of swapping engines and admiring how much quieter they are.
Why does turning data into decisions matter more than the data itself?
I have never been entirely comfortable with the line that data is the new oil. Oil is valuable the moment you extract it. Data, left alone in a warehouse, is close to worthless. It only becomes valuable once someone or something turns it into a decision. The advantage was never the data. It was always the ability to convert information into a judgement worth acting on, and most organisations, if they are honest, have been fairly mediocre at that for a long time.
What AI changes is the cost and speed of that conversion. For the first time, an organisation can put real analytical and decision-making capacity in the hands of a junior manager in a regional branch, not just the handful of senior people who used to carry all the institutional context in their heads. That sounds like a technology story. It is actually an organisational design story, because so much of how companies are structured exists to compensate for the fact that information used to travel slowly and expertise used to sit in a small number of skulls.
Departments grouped expertise together because specialists were scarce and needed to be near each other. Management layers existed largely to move information up and decisions back down, because no one person could hold the whole picture. Neither of those constraints has fully disappeared, but both are loosening faster than most org charts have noticed. Which raises a set of questions I don’t think enough boards are asking with any real seriousness: how many management layers does a business actually need once information moves at network speed rather than meeting speed? Does planning still need to run on a quarterly clock if the environment it’s tracking is changing weekly? I don’t have universal answers to those. I do think the organisations willing to sit with the discomfort of asking them properly will end up with a real structural advantage over the ones who don’t.
What should AI’s role in an organisation look like?
A lot of the public conversation about AI in the enterprise is still stuck at the level of copilots and chatbots, tools that make an existing employee somewhat faster at an existing task. Useful, certainly. But that’s the shallow end of the pool. The more interesting shift, and the one I think gets underdiscussed, is agents that don’t just assist a person doing a job but actually do meaningful chunks of the job themselves.
I’ve largely stopped describing this to my own leadership team as “software” at all. It behaves more like a workforce. It can hold an objective, read context, make a call, and improve the next time based on what happened the last time, with a lot less hand-holding than the tools we were using even two years ago.
I’ve stopped thinking of it as software. I think of it as a workforce.
Picture the specialists you already employ: the person watching for fraud patterns, the analyst tracking supply disruptions, the team personalising a retail offer. Increasingly, there will be a digital equivalent doing some real portion of that work around the clock, not as a replacement so much as a colleague that never sleeps and never gets bored of the two-hundredth data point. I don’t buy the argument that this simply erases jobs. Every big technology shift I’ve lived through, going back to the arrival of ERP systems, changed the shape of work more than it eliminated it. What I do think it eliminates is the excuse for an organisation to remain slow, siloed, and dependent on a handful of people who happen to remember how things were done.
The practical shape of the future enterprise, I’d argue, is three workforces operating together rather than one: the human side, which still owns judgement, relationships, and the things that require actual empathy; the automated side handling the repetitive, rules-based work that never needed a person in the first place; and the intelligent, agentic layer sitting in between, coordinating and reasoning across both. Getting real advantage out of any one of these in isolation is fairly easy. Getting the three to work together without stepping on each other is the actual competitive skill, and very few organisations have it yet.
Will future organisations have as many management layers as they do today?
If a meaningful chunk of coordination work can now be handled by systems that don’t lose context and don’t need to be told twice, it’s fair to ask whether we still need the number of management layers most large organisations still carry. My honest answer, after watching this play out across a fairly complex conglomerate, is: probably not as many.
Hierarchy wasn’t an accident of corporate culture. It was a reasonably sensible answer to a real problem: coordination used to be expensive, and someone had to own it. As that expense falls, because agents can hold context across functions in a way no single mid-level manager ever could, the case for quite so many layers gets thinner. That doesn’t mean organisations go flat overnight, and I’d be suspicious of anyone who tells you it does. What it does mean is that the job of a manager changes shape. Less time spent relaying information nobody else could easily get, more time spent on the things a machine genuinely can’t do well: setting direction, building trust across a team, deciding what’s actually worth doing versus merely urgent.
The organisations that get this right end up smaller at the coordination layer and faster everywhere else. Decisions stop waiting for the next steering committee. Experimentation gets cheaper because fewer people need to sign off before something can be tried. That’s the real dividing line, in my view, between a company that has adopted some AI tools and one that has actually become AI-native: the first bolts intelligence onto the org chart it already has, the second lets the org chart follow the intelligence.
How is the relationship between software and service changing?
There’s a second, quieter shift running alongside all of this, and I think it matters just as much even though it gets far less airtime. For most of my career, software was something a business bought to help deliver a service. A bank licensed a core banking platform and then hired people to run the branch around it. A retailer bought a point-of-sale system and staffed the floor to make the shopping experience work. The software supported the service. People delivered it.
That line is blurring, and in some cases it has already disappeared. An AI agent that can underwrite a straightforward loan end to end, or resolve a customer complaint without ever routing it to a human, isn’t a tool helping someone deliver a service anymore. It is the service. The bank isn’t buying a system to make its staff faster. It’s deploying a system that talks to the customer directly, with a person stepping in only where real judgement or empathy is needed.
This has a knock-on effect that I don’t think gets nearly enough attention: pricing. Software has traditionally been priced by seat or by usage, because the underlying assumption was that a person sat behind it and used it. A service has traditionally been priced on the outcome delivered, or on the labour it took to deliver that outcome. When the software becomes the service, those two pricing logics collide, and the sensible answer is usually to price on the outcome: the claim resolved, the loan settled, the customer kept. Enterprises that keep pricing AI like a seat-based licence are going to underprice it badly, and probably under-staff and under-govern it too, because they haven’t recognised what they’re actually selling anymore.
Where does trust fit into the equation?
None of the above works, at any scale, without trust. And trust is exactly where most of the AI conversation I hear in boardrooms goes quiet.
Older waves of enterprise technology mostly automated things you could see: a transaction, a record, an entry in a ledger. AI increasingly automates things that require judgement: a recommendation, a decision, an action taken on someone’s behalf without them watching it happen. That’s a fundamentally different category of risk, and it deserves a fundamentally different level of seriousness in how it’s governed, not a line item in next quarter’s audit committee pack.
I’ve come around to thinking of trust less as a compliance obligation sitting next to the AI-native enterprise, and more as the operating system underneath it. Everything else runs on top of it. Take it away and autonomy stops being an advantage and starts being a liability waiting to surface at the worst possible moment. Practically, this means building trust in at three levels rather than treating it as one box to tick: technically, so a model’s decisions can actually be explained after the fact; organisationally, so it’s clear in advance what an agent can decide alone versus what it must escalate, and who owns the outcome either way; and culturally, so speed never quietly becomes a substitute for judgement, human or otherwise.
Trust isn’t a governance line item. At this scale, it’s a board-level strategic asset.
One thing I’d add here, and it’s a point I don’t hear often enough: regulated industries have an edge most people overlook. Banking, insurance, and healthcare have spent decades building the muscle for governed decision-making, audit trails, segregation of duties, model risk committees, the lot. Those disciplines were built for human decisions, but they translate to agentic ones better than you might expect. Businesses outside those regulated sectors, moving fast and governing light, would do well to borrow that discipline early rather than learn it the expensive way, in public, after something has already gone wrong.
What is leadership’s role in the AI-native organisation?
Every major transformation I’ve been through eventually arrives at the same point: the technology stops being the hard part, and leadership becomes the hard part. Cloud was never really about servers. It was about how a business thought about capital and risk. Digital was never really about apps. It was about redesigning the customer experience underneath them. AI is not going to be any different in that respect, and it will not, in the end, be about which model you picked.
It will be about leadership, and a fair amount of it lands squarely on the CEO’s desk whether they want it there or not.
What’s different this time is that the mandate has moved beyond sponsoring a transformation programme and tracking an adoption dashboard. There are, I think, three decisions a chief executive now has to actually make rather than delegate. Where does decision-making authority genuinely move to a system, and where does it stay deliberately human, function by function, rather than as a single blanket policy someone can hide behind? How many layers of management does the organisation truly need once coordination stops being the scarce resource it used to be, and does leadership have the nerve to act on that answer rather than just admire it in a slide? And how does the organisation talk to its own people about all of this, honestly, so the shift reads as capability rather than threat.
None of those three can be handed off to a consultant or a task force. They require someone at the top to actually hold a view and say it out loud, then be willing to revisit it as the technology and the regulation and the workforce keep moving underneath them. Boards that wait for certainty before deciding on any of this will find, I suspect, that their competitors didn’t wait, and the gap that opens up is not an easy one to close later.
It will take leaders willing to question assumptions that have shaped their organisations for a generation or more. It will take boards willing to rethink governance for a world where more decisions are machine-assisted. And it will take executives who can hold innovation and trust, automation and humanity, and intelligence and accountability, all at the same time, without quietly dropping one to protect the others.
Above all, it takes leaders who understand that this isn’t another line item on the technology roadmap. It’s the beginning of a different kind of enterprise altogether, one where intelligence sits inside every service, every workflow, every decision, rather than watching from the sidelines and occasionally being consulted.
Technology stops supporting the business from a respectful distance. It becomes indistinguishable from the business itself.
The real question isn’t whether the AI-native enterprise is coming. It’s already here. The question is whether your organisation is building for it, or still waiting for a certainty that is never going to arrive.
That, to me, is what actually separates the AI-native enterprise from everything that came before it. And I’d say we’re still closer to the beginning of that redesign than most boardroom conversations currently admit.
What is your advice to the boardroom?
So let me stop being diplomatic for a moment, because I think the moment calls for it.
If your board has discussed AI only in the context of a technology budget line, you are already behind, and the gap is compounding monthly, not annually. This cannot sit inside the chief information officer’s report as one agenda item among twelve. It belongs at the same table as capital allocation, succession, and risk appetite, because that is genuinely what it now is.
Stop asking your management team for an AI strategy. Ask them a harder question: which of our services could, in principle, be delivered entirely by software within three years, and what does our organisation look like on the other side of that? If nobody in the room can answer, that is your first finding, not a reason to end the meeting.
Do not wait for a perfect governance framework before you move, and do not move without any governance framework at all. Both instincts feel responsible. Both will cost you. Fund the trust architecture, the model risk oversight, and the escalation paths at the same time you fund the pilots, not eighteen months after them once something has already gone wrong in public.
Ask your CEO directly what they intend to do differently about organisational layers, not whether they have “looked into AI”. Looking into it is no longer a credible answer at this stage. Ask what has actually changed in how the business is structured, and if the honest answer is nothing yet, put a date on when that changes.
Protect the people whose judgement, relationships, and creativity your customers will still pay a premium for in ten years, and be equally honest with the roles that will not look the same. Employees can tell the difference between a leadership team that is being straight with them and one that is managing the message. Choose straight. It is cheaper in the long run, and it is the only version that keeps your best people in the building.
Your competitors are not waiting for your board to feel ready. Neither should you.
This is not a five-year plan. It is a decision your board can make differently at your next meeting, and the organisations that make it now will simply be operating a different kind of enterprise than the ones who wait for certainty that, as I said earlier, is never actually coming.