There is a moment most founders hit around now. You have read the articles, watched the demos, listened to enough podcasts on the train. You have seen what some of your peers are doing, sometimes loudly, sometimes very quietly, and you have decided you cannot sit this one out. So you walk into your next leadership meeting and you say it: "We need to build AI."
Then the room goes quiet, because nobody actually knows what that means.
Most of the advice on the internet does not help you here. It is written for enterprises. Six-figure consulting engagements, multi-year programs, governance frameworks designed for businesses with a CIO, a Chief Data Officer, and an eighteen-month runway to "transform." If you run a company between fifty and five hundred people, that playbook is not just unhelpful. It is misleading. It will have you spending money on the wrong things, in the wrong order, with the wrong people, while your nimbler competitors compound past you.
What works at your size is less complicated than the enterprise version, but it requires more discipline. The frameworks are simpler. The decisions are harder.
The first decision: apply AI, or rebuild around it?
There is a version of becoming an AI company that is mostly comfortable. You buy your team Claude licences. You run a few workshops. Maybe you hire an "AI Lead." You bolt some tools onto your existing processes and you get measurable but isolated gains. Customer service responds faster. Marketing produces more content. Developers ship twenty percent more code.
This is what most companies are doing, and it is not nothing. But it has a ceiling, and the ceiling is lower than founders realise.
The reason is structural. You are applying AI on top of an operating model that was designed for a world without it. Your org chart is the same. Your processes are the same. Your hand-offs, your meetings, your release cadence, all the same. You have just made the existing thing faster. The intelligence of the company has not increased; it has been locally accelerated.
“The choice is not which model to use or which vendor to pick. It is whether you are optimising what you already do, or redesigning what you are.”
- Derk Disselhoff
The AI value curve
If you take one diagram from this article, take this one. Most companies plateau at stage two and never realise it.
Stage one is where almost everyone starts. You buy access to GPT or Claude. Your team plays with prompts. Some of them get faster at writing emails or drafting slides. Fine, but it is a productivity feature, not a strategy.
Stage two is where AI gets embedded inside specific workflows. Support drafts responses with it. Marketing generates content with it. Engineering writes code with it. Each function gets visibly faster. Each gain stays inside the function. The org chart, the processes and the hand-offs are still the same as they were a year ago, just running a little faster.
Stage three is where the compounding starts. You build a platform - sometimes called a "company brain" - that connects AI to your actual context: your data, your tools, your customers, your processes, your history. Anyone in the company can build automations, agents, and tools that use that context, without filing a ticket and waiting six weeks. Every new thing built makes the next one cheaper.
Stage four is where the operating model itself is redesigned. The org chart looks different. The roles look different. The way work flows looks different. The shape of the company starts to match the way work actually gets done in a world where intelligence is cheap and abundant. This is what "AI-first" actually means, and it is much rarer than the way the phrase gets thrown around.
The leap from stage two to stage three is the one most founders never make. They keep adding tools, running workshops, producing isolated wins. The curve flattens, slowly, while the team congratulates itself on how much faster everything is now.
Where to focus: the muscle, not the wins
The single most important reframe at this stage of your AI journey is this: stop hunting for quick wins, and start building the muscle.
It is more satisfying in the short term to point at a use case that saved someone three hours a week. It looks good in a board update. It gives the AI program a story to tell. The problem is that three hours a week, fifty times over, does not make a company AI-native. It makes a company slightly faster at being what it already was.
The muscle is the underlying capability that lets you build the next thing, and the next, and the one after that, faster every time. In practice it looks like four things:
- 01A platform where AI is connected to your real business data with proper access controls.
- 02A growing library of reusable skills, prompts, agents, and connectors that anyone in the company can pick up.
- 03A cadence of training and sharing - typically a weekly demo session - so that what one person builds becomes available to everyone else within days.
- 04A culture where building something with AI is the default response to a problem, not an exception that requires permission.
“A team that has saved a thousand hours through fifty disconnected automations is not the same as a team that has built one platform on which a thousand automations run. The first is a productivity story. The second is a company.”
- Derk Disselhoff
Do not build a separate AI team
This is counterintuitive, and it is the single most common mistake founders make. When you decide to take AI seriously, the instinct is to create a dedicated team. Hire an AI Lead. Give them a couple of engineers. Place them somewhere on the org chart where they can move fast without being slowed down by the rest of the business. The intent is right. The structure is wrong, and it fails for reasons that are almost always invisible until twelve months in.
A separate AI team kills the thing that makes AI valuable, which is context. The people who understand how the business actually works - the customer service lead who knows why refunds get tricky, the operations manager who knows which suppliers are slow, the merchandiser who knows which products sell badly in November - those people are not on the AI team. The AI team, working in isolation from this context, ships things that either solve the wrong problem or take so long to solve the right one that the moment has passed.
What works instead is embedding AI capability inside the existing teams. The customer service team builds their own AI workflows. Operations builds theirs. Product builds theirs. There is a small central function, often one or two people, whose job is to enable these teams, not to do the work for them. The central function looks after the platform, the standards, the data plumbing, the security posture, the training cadence. The teams own their use cases and the outcomes attached to them. Context stays where it lives. Capability scales without bottlenecks.
Who to hire: the new competency stack
The competency framework for new hires has changed, and most founders have not updated theirs. The single most useful filter at the door is whether someone is what you might call an "AI-fluent business thinker." Three things sit underneath that label.
- 01They understand the full business context. They can hold the customer, the operations, the financials and the product in their head at the same time, instead of being a specialist who can only solve their own slice.
- 02They have an extreme growth mindset. They are comfortable working without structure, without precedent, without a manual - because increasingly, there is no manual.
- 03They are obnoxiously curious. Not "interested in technology" curious. The kind that asks why three times in a row, the way a small child asks why the sky is blue, until the adult runs out of answers.
The last one is the hardest to test for in interviews and the most valuable when you find it. Hire for the people who keep asking why.
The whole thing on one page
Not a transformation program. Not a consulting engagement. A founder, an enabler, a weekly session, a one-page strategy, and one honest number to track.
The companies that win the next five years will not be the ones that ran the biggest AI programs. They will be the ones that started the muscle-building earliest and protected the rhythm of it longest.
“The good news for you is that this does not take an enterprise budget to start. It takes a decision.”
- Derk Disselhoff
