Where did the idea come from?
It started in marketing, the way it probably did for a lot of people, only earlier. Before AI went mainstream I was writing my first texts in OpenAI's Playground, for marketing pieces, assets and whatever else we needed. Some I wrote with it, some it drafted for me, and from a very basic starting point there was not much left to do. The texts still needed improving, and everything was rudimentary.
That was the first turning point for me. Boring, repetitive processes could be automated, and above all repetitive tasks: LinkedIn posts, copy, articles.
Then the first really good automation builders arrived, n8n and Make, well before OpenAI or Anthropic could connect to other systems. Everything went through APIs and took a lot of hands-on work. That is when I noticed I could build input forms with context, send them to the AI, and get back assets and content for a given input, goal and channel: texts, images, everything that mattered in marketing. Building websites or landing pages with AI was still a thought about the future.
That was when the idea came: by 2026 or 2027 at the latest, we would be able to automate core tasks all the way through, and the person would only have to say release, iterate or update.
What came true, and what surprised me?
More came true than I expected. We are further along than I thought we would be.
The part that was not really on my radar was the tools a company already uses connecting directly to AI. I knew more and more vendors were opening their APIs. I had not really pictured a connection where I sign in with my own account, so the AI has exactly the rights I have in the tool and none that I do not. A self-built server on one shared API key gives everybody who reaches it every function the key allows, and that carries a certain security risk.
What is AI, if it is more than a chat window?
Most people met AI as a chat window, so that is what they think it is: something you type a prompt into. That first picture is the main problem with how AI gets used.
I see it in three roles. It automates. It is a sparring partner. And above all it carries out the work.
I never stopped prompting. The way I prompt changed. The old best practice gave the AI a role, such as an experienced performance marketing manager, then a goal, the task itself, the steps along the way, what it must never do, and what the output should look like. I have not prompted like that in months. All of that is now replaced in the system by a simple prompt in plain language, because agents in the system break that prompt apart. What that buys is reuse, and a prompt quality that stays the same.
So today I say: this is the vision, this is the campaign we want to run, here is the knowledge and the briefing, this is the goal, go. Then the result gets feedback, we iterate, and we keep going until the output is what the person wants.
You could almost say it has become a kind of human conversation with a machine, with all that comes with it. I can tell it that a draft is rubbish. I can tell it that a draft is excellent. And I can tell it to take what worked into the rules it works from, as best practice, so that given the right input in the right setting, it draws on that best practice next time.
The test for whether a company runs on AI has three conditions, and I wrote them down in what an AI-first company is.
What does an AI-first company look like day to day?
In three years, the way I see it, the vision becomes the input. A marketer will no longer walk in thinking "today I have to do X". The team meets, agrees on a vision for the next two weeks, and the marketer puts that vision into the system. The system turns it into a first draft that is close to ready, from everything it is connected to: web analytics, HubSpot, social media data and the rest.
Whether that still happens in a chat window or somewhere else is open. It will work in a similar way.
What decides the result is what the person brings: the creativity, the knowledge, and the vision as it first exists in somebody's head. The more detailed and the more creative that input is, the better the result.
Inside a B2B GTM organization, we built a campaign process where one campaign goal goes in, with its context and a briefing. The system builds the social graphics and the LinkedIn and Meta ads with their images, and for an event the landing page and the blog post. When something is missing from the brief, it asks. The person says yes, no, or what needs to change. The vision and the creativity stay with the people, and the execution sits with the system.
From six months inside a B2B GTM organization with 36,000 accounts in its CRM.
For a salesperson part of it is already simpler: the day starts with a briefing on the accounts assigned since the previous day, with the contacts already researched. I built that as a routine that starts on workdays at 07:30.
What we have only just started there: an AI-augmented SDR setup, so the salespeople no longer have to write their emails themselves. In three years I expect the emails already prepared from the knowledge base and waiting to be sent, and accounts and contacts researched in far more depth than today. What is left then is picking up the phone.
The customer will change too. The personalization I mean goes past mentioning somebody's latest post. The technology will bring buyer and seller together, in a certain form, before the first conversation. And the buyer will know much better what their situation needs: which tools, and which services.
So the question is less what somebody does all day, and more how a company's vision gets done, together with the creative people behind it. People will work with each other far more, as a team, to get these things done. And they will work with the machine far more, to turn that into a result.
What has to exist before a company can become AI-first?
From a marketing view and from a sales view, what is missing above all is a written description of what the company does. Its positioning, its offering, its messaging. How it presents itself, its brand design, its colors. Everything that makes a company a company. Information about existing customers matters too. Very important, of course, is the ICP, meaning who the company sells to, and the personas.
If none of that is written down, an AI-first organization is hard to build at all. Those foundations have to exist, or be created first, before anything worth that name can come out.
Everything else, a campaign process for instance, comes out of the processes and tasks a company already has. It comes out of daily life in a company, never out of a wish list.
What did I get wrong?
A lot, because this was unexplored ground. I had to learn fast how AI works, how it does not, and how to work with it to get what I wanted.
I wrote texts that did not sound human. I published things that were weak in substance. I relied too much on the AI without asking whether the output was any good. Those are the growing pains, the kind you only get by doing it.
I also got the way I connected AI to other systems wrong. An agent has to know which tool it may use for which question. When I asked which recent discovery calls with prospects had gone well, it went into HubSpot as well, which was outside the scope of the question.
Not every failure was a connection. A scheduled routine of mine once failed for six working days without reporting an error anywhere, because the file holding its instructions had been deleted.
Cost was a reason too to limit which tool an agent uses. The usage ate into our weekly allowance, and sometimes we could not keep working after two or three prompts. Which model for which job, how much effort for which task, when a small model is enough: all of it learned by doing. Once I ran two prompts on Claude Fable, and my daily quota was used up for the next five hours.
Why will a company that does not run on AI fall behind?
Something will still work without AI. The question is why a company will not be able to hold its ground without it in three years, and the answer is fairly simple. Its competitors will be faster, more effective and more efficient, and over time they will take its customers. I think a few young people will arrive with the same idea for a tool as an established company, and leave it behind.
The point is the same team delivering far more, and possibly better, because it no longer depends on general knowledge it used to hire for. Replacing people has nothing to do with it. I can set up a Meta or LinkedIn ad campaign today without much knowledge and without ever having a performance marketing manager. Whether those ads perform is another question. But how to build them is more or less common knowledge, and it no longer sits in a few people's heads.
Maybe it is simply this: a company cannot afford the next hire, say at 60,000 euros a year gross. It can start with an AI plan from OpenAI or Anthropic from about 20 euros a month. A model may not be as creative as a person, and with it I get the first version done faster. That is about the next hire, never about shrinking the team.
The tools are going AI-first. If I am not a HubSpot expert and want to know what is happening in the CRM, I can ask the AI connected to it in plain language instead of clicking my way through. If I want a process automated in Clay, I can have it built through the command line, test it once, and if it runs, it runs.
Many tools are going headless: once they are connected to the AI, I can work with them entirely in natural language and get what I want without ever opening the tool.
This piece started the same way. I answered the questions behind it by talking into a voice-to-text app.
The technology also decides which skills a person needs. People who flatly refuse to work with it will, over time, be replaced by people who want to. That is the natural course of things.
Picture two companies side by side for the same four weeks. In the first, a two-hour meeting on Monday sets the vision for the coming weeks, using what the data from its tools says, and the work goes straight into execution with the creativity of the people behind it. In the second, the team sits down and starts working out which assets it needs, which processes, and what else has to be done. Then somebody is out sick, and somebody else is not there, and all of it blocks.
At the end of the four weeks the first company has results, maybe not the right ones and maybe not good enough yet, but it has them. The second is still stuck in execution. So which one gets more customers? Which one can grow, and which one shrinks?