The AI organisation: why most companies are doing it wrong
I work with companies trying to figure out AI. Big ones, small ones, somewhere in between. And after spending time inside these organisations, seeing how they operate and where they get stuck, I can tell you that most of them are doing it wrong. Not because they don’t want to use AI. They do. But because they think buying subscriptions is the same thing as transformation.
It’s not. Not even close.
The subscription trap
Here’s what happens at most companies. Someone in leadership reads about AI, gets excited, and rolls out a ChatGPT Enterprise licence or a Cursor subscription to the team. Maybe they run a workshop or two. Maybe they send around a PDF with prompting tips. Then they move on to the next quarterly priority and assume the job is done.
And sure, it helps a little. People write emails faster. Developers autocomplete some code. Someone in marketing generates a few blog posts. The tools are good and they do speed things up in isolated moments.
But nothing fundamental changes. The organisation still operates the same way. The same meetings happen. The same approval chains exist. The same number of people sit in the same roles doing roughly the same things, just with a slightly fancier text editor. It’s like giving everyone a sports car but keeping the speed limit at 30.
Some companies take it one step further and introduce an “AI day” once a month. A dedicated day where everyone is encouraged to experiment with AI. It comes from a good place but the effect is basically the same as buying subscriptions. It signals that AI is something separate from real work. Something you do on a special day, like a hackathon or a team-building exercise.
But AI shouldn’t be an event. It should be how you work every single day, on every task, as naturally as opening your laptop. Having an AI day once a month in 2026 is like having an “internet day” in 2005. If you need a dedicated day to use the most transformative technology of our time, you’ve already missed the point.
Then there’s the companies that appoint a “Head of AI” or “AI Director” to lead the transformation. It sounds serious and strategic but it’s the same mistake in a different costume. You’re taking something that should be embedded in how everyone works and turning it into one person’s responsibility.
It’s like hiring a Chief Internet Officer in 2003. The internet wasn’t a department. It was the foundation everything else was built on. AI is the same. The moment you isolate it into a role or a team, you’re telling the rest of the organisation that AI is someone else’s job. And they’ll happily let it be.
The tools are not the bottleneck. The organisation is.
The walls AI keeps hitting
When you start using AI seriously, not just as a writing assistant but as a way to actually build and ship things fast, you quickly run into walls. And those walls are almost never technical. They’re organisational.
Approval processes that require three people to sign off on a colour change. Deployment pipelines that only certain teams are allowed to touch. Role definitions so rigid that a designer building working code is seen as a threat rather than a breakthrough.
Meetings that exist to update people who exist to attend meetings. Decision trees where every choice routes through someone’s calendar. Tools that don’t talk to each other because different departments chose different vendors five years ago.
These structures were built for a world where work was slow and people were the primary bottleneck. You needed people to route information, align teams, manage dependencies, and keep things on track.
But AI changes the equation. The work itself can now happen incredibly fast. I can build something in two days that used to take months. The bottleneck is no longer doing the work. It’s everything around the work. And most organisations have a lot of everything around the work.
Two stories from the field
I’ll keep the names out of this, but these are real situations I’ve been involved in recently.
The first is a medium-sized company with a sizable development team. When we started using AI to build front-end features directly in code, skipping the traditional handoff from design to development, some of the developers got nervous. New processes appeared almost overnight. Code reviews that hadn’t existed before. Architectural approval steps for simple UI changes. Requirements for documentation that nobody had previously asked for.
Some of it came from a genuine place. Quality matters. Standards matter. But the more AI proved it could produce solid work quickly, the more the resistance grew. It started to feel less like quality control and more like territory protection.
The result was that something AI could build in an hour took weeks to get through the process. I’ve written about this shift in designers who ship and it’s happening everywhere. The roles are blurring and not everyone is comfortable with that.
The second company is based in Silicon Valley. Different culture entirely. They looked at what AI could do and made a radical call: they let go of their entire front-end development team. Product managers and designers would build directly with AI from now on. No handoffs, no tickets, no waiting.
Is it too soon? Maybe. Will the quality hold up? We’ll see. But the speed at which that company now ships is remarkable. And they’re not alone in thinking this way.
What’s happening around the world
This isn’t just happening in small pockets. It’s a global shift and it’s accelerating.
Oracle laid off up to 30,000 employees on March 31st with a 6am termination email. That’s roughly 18% of their workforce. The stated reason: freeing up $8 to $10 billion to fund AI data centres.
The company posted a 95% jump in net income last quarter. This wasn’t a company in trouble cutting costs. It was a profitable company deciding it needs fewer people and more compute.
Klarna went hard on AI early. Their CEO Sebastian Siemiatkowski announced that an AI chatbot was handling the work of 700 customer service agents. They cut their workforce from about 5,500 to around 3,400, a 40% reduction.
Then he admitted they went too far. Quality dropped. Customer trust eroded. They’re now rehiring humans for the roles that AI couldn’t fully handle. A cautionary tale about moving too fast without understanding what you’re cutting.
Shopify’s CEO Tobi Lütke sent an internal memo that leaked and then he shared publicly. The core message was blunt:
“No one in the business is allowed to hire a human without first proving that AI cannot do the job.”
He added that AI usage would be part of performance reviews and that this applied to everyone, including himself and the executive team. Shopify went from about 5,000 to almost 3,000 employees.
Lütke claimed employees could use AI to “get 100X the work done.” Whether that number is real or aspirational, the direction is unmistakable.
Jack Dorsey at Block outlined a vision where AI takes over the coordination tasks traditionally handled by middle managers. Tracking projects, assigning tasks, providing business insights. The roles that exist to manage information flow between people become unnecessary when AI can do that instantly.
Gartner predicted that by 2026, 20% of organisations will use AI to eliminate more than half of their current middle management positions. That’s not a fringe prediction from a startup blog. That’s one of the most established research firms in the world.
And on the other end of the spectrum, solo founders are on the rise. According to Carta, 36% of startups founded in 2025 had a solo founder, up from 24% in 2019. Midjourney generates roughly $4.7 million in revenue per employee. Traditional SaaS companies average $200,000 to $300,000.
Dario Amodei, CEO of Anthropic, estimated a 70 to 80% probability that a billion-dollar company run by a single person will emerge in 2026.
It already happened. Sam Altman predicted in 2024 that a one-person billion-dollar company “would have been unimaginable without AI, and now it will happen.” He recently emailed the New York Times saying he won a bet with tech CEO friends over when it would arrive.
The guy: Matthew Gallagher, 41. He spent $20,000 and two months building a GLP-1 weight-loss telehealth company from his living room in LA. His stack: ChatGPT, Claude, and Grok writing code. Midjourney for images. Runway for video ads. ElevenLabs handling customer calls. Custom AI agents stitching it all together. $401 million in revenue in year one. On track for $1.8 billion this year. One person.
The leverage AI gives small teams is unlike anything we’ve seen before.
The small team advantage
I’ve experienced it firsthand. I can build and ship things in days that would take a traditional team weeks or months. Not because I’m faster than other people. Because AI removes the need for most of the coordination, handoffs, alignment meetings, and waiting that slows large teams down. When one person has the context, the tools, and the agency to make decisions and execute, everything moves at a different speed.
The key is not just using AI. It’s removing everything that slows AI down. Small teams have always been faster. AI makes the gap absurd.
What the AI organisation looks like
The most effective companies in the AI era will look nothing like today’s organisations. Fewer people. No middle layers. No managers whose only job is managing other people. No juniors learning on the job while seniors review their work.
Instead: a small group of high-agency people, each with deep expertise in something, but all working broadly across the problem space. They’ll orchestrate fleets of AI agents to handle everything from research to code to strategy to communication. Their job will be to think, decide, and direct. The AI does the bulk of the execution.
Everyone contributes real work. Including the CEO. If your job is attending meetings and making PowerPoint decks that summarise what other people did, that job is gone. The AI can summarise, align, and report faster and better than any human middle layer.
Strong leadership still matters. Maybe more than ever. But leadership in the AI organisation means being hands-on, building alongside the team, not overseeing from a distance and reviewing what others built last sprint.
That’s the upside. But there’s a harder side to this story that I don’t want to gloss over.
Hela havet stormar
There’s a Swedish children’s game called “hela havet stormar.” You might know it as musical chairs. Everyone runs around and when the music stops, there aren’t enough seats for everyone. Someone is out.
That’s what’s about to happen in the job market. And I don’t think people fully grasp it yet.
The examples above are just the ones making headlines. Across every industry, companies are realising they can do the same work with far fewer people. And once that realisation sets in, the chairs start disappearing.
The uncomfortable truth is that many of these jobs are not coming back. AI doesn’t just automate manual tasks. It automates thinking, writing, coding, analysing, designing, coordinating. The kind of work that entire white-collar industries are built on.
Some companies will move too fast and have to backtrack, like Klarna did. Others will move too slowly and lose to competitors who figured it out. There will be a messy period where everyone is trying different approaches and nobody knows the right answer. That period has already started.
But through the chaos, one thing is becoming clear. The people who will thrive are the ones who adapt. The high-agency generalists who can use AI to do the work of ten people. The ones who don’t wait for a manager to tell them what to do. The ones who learn fast, ship fast, and take ownership of outcomes rather than tasks.
The gap between those who work with AI and those who don’t is already wide. It’s about to become a canyon.
The only way forward
I don’t think there are multiple valid strategies here. Companies can disagree on the pace, and they should, because moving too fast breaks things and moving too slow kills you. But the direction is the same for everyone.
Remove the walls. Flatten the hierarchy. Give high-agency people the tools and the mandate to move fast. Stop protecting roles that exist only because they’ve always existed. Stop adding process to manage what AI makes trivial. Stop treating AI adoption as an IT project and start treating it as an organisational redesign.
And here’s the part that most companies miss: don’t reorganise for what AI can do today. Reorganise for what it will be able to do in six to twelve months. Things that AI struggled with last quarter it handles effortlessly now. If you design your organisation around today’s capabilities, you’re already behind by the time you’ve finished implementing the change. Plan for the trajectory, not the snapshot. What feels like a bold bet today will feel obvious in six months.
The companies that win will be the ones that tear down the obstacles between intent and execution. Not the ones that hand out subscriptions and hope for the best.
And if you’re an individual reading this, the same applies to you. Don’t wait for your company to figure it out. Don’t wait for permission. Don’t wait for the perfect tool or the perfect workflow. Start building. Start shipping. Start proving what’s possible with AI, even if the people around you aren’t ready to hear it.
The music is playing. But it won’t play forever.
