Treat AI like electricity, not a killer app

One big thing: AI is a power source, not an app

The real problem isn’t tokens or broken pilots or galactic-level cost. It’s that we’re treating AI like an app—an IT problem—instead of what it actually is: a new power source poised to drive an operating model shift across nearly every company that uses data, software, and people to do work.

The last time we faced something this powerful was when we electrified our homes, factories, and offices. Before that, we used water wheels: lovely machines, effective for centuries, but limiting in myriad ways.

Then came electricity, and we had to reimagine how work got done, how value flowed, what skills we needed in the workforce, how we even build a factory or an office. Electrified businesses built the next century. The ones that treated electricity as a better way to turn the old water wheels – and there were many – got left behind.

That’s AI. AI gives us a new power source around which we can – must – rewire how our companies generate value.

This is not consulting AI-washing; it’s what data and experience are telling us.

We’re being poached in the assertion that AI is a “problem” for corporate IT to solve. (Nope.) Or that a 10% productivity improvement is worth the time and cost and risk. (It’s not.) Or that if one waits out a business cycle, things will work themselves out. (Carefully do the maths on your retirement funds.)

Untying this knot to make informed real-world decisions requires clarifying the line between infrastructure – procurable from outside the firm – and core – what your company does that is truly different and value generating. (If this sounds like echoes of Hammer and Champy from the early 1990s, good ear.)

This problem is exacerbated because enterprise IT has spent the last seventy years doing the opposite: building and managing infrastructure inside individual companies. We layer in ERP systems like Salesforce, Oracle, ServiceNow. Build or lease data centres. Wrestle with cloud strategies. Manage PCs, networking, ping, power, pipe.

It made perfect sense at the time, but we’ve held onto the 1960s model of bespoke personalised technology infrastructure even as the world moved on. Our cloud journey from the late 1990s illustrated this as companies large and small fought against centralised cloud compute: it’s too expansive, not secure, not auditable, and so on. In the early days, that had some truth, but – let’s be honest – those hurdles got lowered quickly. In 2015, Nicholas Carr saw this trend and compared corporate IT to electricity and predicted its end. That hasn’t happened (and won’t), but enterprise IT is now certainly poised to be dramatically re-shaped by AI; the question is: how fast?

When OpenAI’s technology left the lab way back in November 2022, corporate IT leaders did what they know how to do: start building your own AI factory; over-investing in bespoke infrastructure to harness a revolutionary technology. Then we all bought licenses, held workshops, did the pilots, clutched our pearls over the doomsday future. And then, most didn’t get the hoped for results.

If we look at history, those “failed” early experiments were entirely expected. The reassuring news is that this period of experimentation and invention, based on developing competitive advantage, has been throughout history a good thing. It turns out that we need that running start to leap onto new S-curves of growth. It happened with steam, locomotives, electricity. It probably happened with fire. It’s happening with AI right now.

What comes next is also predictable. A wave of innovation as companies realize they don’t need to own the infrastructure. The smart money is on figuring out how to use the power generated and distributed from outside the walls of the firm to profoundly disrupt how work is done inside your firm, with your people, your processes, and your existing technology.

Five no-regrets steps to take tomorrow

1. Don’t just electrify your water wheel

When electricity came online in the late nineteenth century, we didn’t have centralised infrastructure. So thousands of companies sensibly built private power stations. A head of electricity used to be a real job. Eventually we figured out that electricity generation and distribution worked best as infrastructure—centralised, metered, reliable. It got to where we could simply plug in a typewriter or blender or radio. It just works, and you pay for the watts you use. Most of us don’t need to build our own power plants.

Unless you’re in the infrastructure business, your job is not to manage electricity. Or telephony. Or—now—your entire AI stack. Your job is to use that power for what makes you distinctive, and then stop competing over non-core infrastructure. (Usman Sheikh calls these “rails.”)

To be fair, it’s often not that easy to recognize your true core. Whether you’re a bank, a pharma company, a healthcare insurer, a telecom, or an educational institution, the work may have been done well the same way for years or decades.

And that work still needs to happen, but we should break from the assumption that every company needs to own all their own AI infrastructure. (How many companies lost billions trying to create their own cloud offerings only to get smacked by Amazon, Microsoft, and Google? Ouch.)

Early winners, like the examples above, are starting at the workflow level: Software engineering, HR, IT operations, or business processes that are high-friction, high-cost, routinized, and data-intensive.

Our job now as business leaders is to align AI agents, platforms, and people to the workflows that matter, and demand results from service and platform partners who specialize in taking AI the last mile.

2. Stop. Blaming. The. CIO.

AI is not an app. It’s a forcing function for a new operating model across nearly every firm.

Yes, it involves computers and code, so we’ve defaulted responsibility to the CIO’s desk. But that’s a problem. Treating AI as one person’s job—regardless of title—makes it feel optional, something to place somewhere in the organisational chart under special projects, the Chief AI Officer, the beleaguered CIO/CTO.

AI is too big for any one role in a major enterprise. Real traction happens when the entire C-suite owns the shift. Every company will have champions and detractors, and that’s healthy, but the point is that the impact AI offers is too pervasive to be any one person’s job alone (even the CEO). Culture, leadership, execution rigor, outcome discipline are now board-level concerns, not just another monkey on the backs of the enterprise IT team.

If half your C-suite feels like “Value from AI is not MY problem…,” polish your resume.

3. Embed trust in your AI operating model

AI is augmenting cognition itself, and that triggers something primal. Instead of “fire = warm food and safety,” our brains scream “fire = danger!” That’s one reason we can’t quite trust AI enough yet. When we don’t trust something that’s starting to “think,” we pull back. It’s not just a technology shift. It’s a psychological shift that will precede disruptive value from AI.

That’s the AI Trust Valley. And just like in any valley, the only way out is through. Fortunately, we’re learning how to build trust into AI-enabled work.

It starts with treating agents and governance as a foundation, rather than a speed bump we may try to drive around.

Nearly every AI programme failure has roots in a governance failure. Demand transparency that can help demystify AI and ensure trusted outcomes. Leave humans at the wheel. We need to get very tactical here because work happens step by step. Figuring out where and how carbon-based humans can work with silicon chips powering AI is the essential algorithm that will lead to success. Silicon chips can do the heavy cognitive lifting, but we need human hands on the wheel at the task level to ensure security, ethics, process control, and accountability.

4. Evolve your ecosystem for the AI operating model

Yesterday’s safe-bet Tier 1 partners may not stay Tier 1 in the AI world. Your ecosystem needs to change, and we’re learning from early winners that this changes procurement, legal, finance (again, not just IT).

Modern partners need to demonstrate humans and AI agents working together across a specific workflow. Some will be horizontal— my own firm Ascendion for software engineering, Salesforce Agentforce for sales and service. Others will be industry-specific: Palantir for banking operations, Harvey for legal work, Abridge for clinical documentation. Thousands more are coming online.

The modern firm is an ecosystem that now must include a constellation of partners who leverage LLM technologies, have tools and processes to take value to the last mile, and contract for outcomes.

5. Contract for impact

Your business metrics from yesterday are, in more cases than not, your business metrics tomorrow. Cost per member per month. Customer acquisition cost. Cost-to-income ratio. Star ratings. Same-store sales. Time-to-market. You already speak this kind of language with your board and investors and your teams.

We know now that pitching Claude licenses into your enterprise and hoping for the best is not going to work. So stop counting tokens as a measure of AI impact—that’s a measure of horsepower, not outcome.

Use your essential metrics to shape contracts related to AI products and services. Demand AI Arbitrage in your pricing to tie AI to outcomes and impact. Contract for the metrics that mattered yesterday, because those are the numbers that ladder up to the CFO. That helps assure business impact (not pilot paralysis), externalizes AI risk, and wires your operating model into your ecosystem.

AI will profoundly disrupt how you move these numbers. But what gets counted changes less than many think.

There’s no free lunch and no free AI. The LLM foundries put an exclamation point on this when they moved from a freemium model to a token model. Nearly overnight, CIOs and CFOs all over the world were slammed with bills that weren’t completely expected without the commensurate business impact.

It’s a painful and necessary reminder that power generation, railroad tracks, and LLM cognitive “force” have to be managed for the future, and that means reframing how we’re managing and accounting for AI within our own companies.

We’re all being simultaneously confronted with AI doomsday naysayers and doe-eyed optimists. The rest of us in the practical middle are trying to do the right things for our companies and colleagues. The path ahead may not be easy, but the blueprint is becoming clear. Step one is to reframe AI less as a pure technology or “app” and more as a power source for modernising nearly every firm.

Paul Roehrig is the Chief Strategy and Marketing Officer of Ascendion. He is also the co-author of: Monster: A Tough Love Letter On Taming The Machines That Rule Our Jobs, Lives, and Future (2021), What To Do When Machines Do Everything (2017), and Code Halos: How the Digital Lives of People, Things, and Organizations are Changing the Rules of Business (2014).