Most guides say the same thing: if you want to break into AI, start an automation agency. For most people, I think a different route is more realistic — becoming the person who knows AI best at the company you already work for. It's a path the market shows ever more clearly, even if it rarely gets talked about. Below I'll break the argument down to its parts.
First, two terms that will keep coming up. An AI automation agency is a company that implements AI-based solutions for other businesses — usually automating repetitive processes. An agent is a program that carries out a multi-step task on its own (reading a ticket, finding the answer, drafting a reply, say) rather than just answering a single question. An implementation is taking a tool like that from idea to everyday use inside a company.
Where these numbers come from
The starting point is an IBM study that surveyed 2,000 CEOs of large listed companies — with a median annual revenue of around $5.8 billion (IBM, CEO Study, 2026). And right away an important caveat: this is about big, established firms, not a global average. What's true for these CEOs needn't be true everywhere. Globally the percentages are probably much lower — I'd be surprised myself if they topped 30 percent.
A new role that appeared in two years
The first number worth pausing on: 76 percent of the CEOs in this study either already have a chief AI officer (CAIO) in the company or are hiring one this year. Two years earlier that figure was 26 percent.
The easiest way to grasp this is by analogy. The CEO runs the whole company, the CFO runs the money, the COO runs operations. There's also the chief information security officer (CISO), responsible for cybersecurity. That last role didn't exist until the internet arrived — in 1980 nobody needed it, because cyberattacks weren't yet a real cost. Once they became one, within roughly fifteen years every larger company had to put someone in charge of it.
Exactly that is happening with AI now — except it didn't take decades, but about 24 months. A new problem appeared that nobody at the top knew how to handle, so a new role was created. And here's the crux: you don't have to become a chief AI officer to benefit from this shift. The same study says that every other member of the leadership team — from marketing to finance and operations — is also meant to become fluent in AI. One chair is the most visible, but there are far more seats open than that.
The gap that keeps CEOs up at night
The second number shows where the bottleneck sits. At the firms surveyed, only 25 percent of employees actually use AI tools in their daily work. At the same time, those same CEOs claim that 86 percent of their people either have the right skills or would pick them up after a short course. That's a 61-point gap.
To be fair: this is survey data, and employees tend to understate what they really use — so the exact size of the gap is uncertain. But the gap itself is real. The reason companies still haven't rolled AI out everywhere isn't that they lack people with the skills. They lack someone to connect those people with the processes that genuinely need it — someone to build the bridge between "we know how to use it" and "it's saving us forty hours a week."
A bridge like that doesn't build itself. Someone has to be curious enough to take the initiative — or be handed the task outright. Few people rush to it, because change management (walking a team through a new way of working) can be exhausting, especially in a large company. First you have to absorb the short-term cost — train people, take existing processes apart and put them back together — before the long-term gain arrives. That up-front pain puts decision-makers off. Sometimes it's wiser not to rework the whole organization, but to spin up a new department from scratch, AI-based from the start, with no old habits to undo.
Two roads to that chair
I see two paths, and one of them, I think, is more accessible than people assume.
Path A is the classic one: you start as a consultant or set up an AI agency, take on a few clients, solve their problems. At some point a few of them decide it's better to put you on staff. You enter the company "from inside," through work you've already done for it.
Path B is internal promotion. You already have a job. Quietly, you're the person who knows AI best in the building — you turn up to meetings with a ready-made prompt that saved the team three hours, you build small automations nobody asked for because you're curious and you experiment. When a new position opens up, you're the obvious candidate.
Most people assume only Path A exists — because that's the story you hear most often. Yet a separate IBM study shows otherwise: of 600 chief AI officers, 57 percent were promoted from within (IBM, study of 600 CAIOs, 2025). They were already on the spot, doing the work, before the title for it even existed. Crucially, neither path requires you to wait until you're hired: you can start closing that gap in your current role right now. Pick one process in your area that no one on the team has touched with AI tools yet, build an AI-assisted version of it, document how much time it saves, and show it to your manager.
Passion decides, not the title
This, to my mind, is the most important piece of the whole puzzle. You can't stick with something you don't enjoy. Many people hear "want to make money in AI? start an agency" and assume they have no choice — even though they have no wish to run sales calls or chase clients and hear "no" fifty times before the first "yes." And that's perfectly fine.
My advice is simple: build an AI-assisted version of what you already do and enjoy. If you like marketing, don't force yourself to build finance agents — automate content, copy, pages. You'll become a marketer fluent in AI, the kind a director eventually promotes. Trying to be someone else is a common source of impostor syndrome — an ailment plenty of people learning AI complain about.
The data from that study of 2,000 CEOs backs this up. 85 percent of them say every functional manager in the company has to become a technology expert — and it's the CEOs saying it, not consultants. 77 percent believe the "people" and "technology" roles are merging: the winner will be whoever bridges both worlds.
One more analogy captures it well. When the internet arrived, there were "internet marketers" and "internet agencies" — a separate category on the business card. Today, if someone introduced themselves as an "internet marketer," it would sound odd, because the internet has soaked into all of marketing and the adjective has dropped away. I expect the same fate for AI: today we have AI consultants and AI agencies, but soon there will just be consultants — and anyone who isn't fluent in AI won't keep up with the rest.
Two honest notes to close
I'll stress caution about these numbers twice over. First, they come from large, established firms — most of us don't work at those, and globally the percentages are probably markedly lower. Second, CEOs can be wildly off in their forecasts: in 2024 half of them thought AI would be the main growth driver by 2026, and today only 10 percent say so — a miss of around 40 points in a year.
What isn't up for debate, though: CEOs are hiring, the structure of companies is changing, and the functional managers getting promoted are the ones who know AI best. One line from the IBM report stuck with me: "Today AI supports people. By 2030, people will support AI. The biggest change won't be structural, but cultural." In other words — the way we work is changing, and the tools follow from that.
Regulated industries deserve a note of their own — healthcare, finance, the public sector — where you can't just plug AI into the company's data. You might think the topic doesn't concern them. I think the opposite is true: domain knowledge combined with AI fluency under constraints is one of the rarest profiles on the market today. If you're hearing "no" for now, build your own projects after hours, on stand-in data, and show them to the team — so that the moment the company gets the green light, your name comes up first.
The conclusion is brief: you don't have to change your profession, you have to change which version of your profession you are. Thinking can be outsourced; understanding can't. So choose your path deliberately.