What is an AI-first company?
An AI-first company runs its recurring work in an AI system until a person only has to judge the result. That is narrower than "we use AI", and it is meant to be, because it can be checked.
Three conditions have to hold at the same time:
| Condition | What it means | What it looks like |
|---|---|---|
| It starts without a reminder | A time, an event or a signal starts the work. Nobody has to think of it | A briefing on each rep's newly assigned accounts starts on workdays at 07:30 and lands in a dated file, before anybody asks for it |
| It runs through | Several steps happen in a row, and nobody nudges the work along between them | One campaign goal and its brief go in, and the system builds the graphics, the ads and, for an event, the landing page. It asks when the brief is missing something |
| A person decides at the end | The person judges a finished result, sends it back with feedback or releases it | Yes, no, or this is what needs to change |
Miss one of the three and what you have is assistance. Assistance is useful. It is the starting point and never the goal.
The campaign process in the table is one we built inside a B2B GTM organization. This is why:
We wanted redundancy whenever someone was out of office for a certain amount of time. And we wanted every manager and every individual contributor in our organization to be able to kick off entire campaigns on their own, without deep knowledge of every part of the marketing organization.
From six months inside a B2B GTM organization with 36,000 accounts in its CRM.
My view, and the reason I build this: in three years a company that does not run on AI cannot keep up with its competitors.
How is running on AI different from using AI?
Using AI means a person starts every task and carries it through, with AI helping at some of the steps. Running on AI means the task starts and runs on its own until a person judges the result.
The Content Marketing Institute asked 1,015 B2B marketers in the summer of 2025, and three of their answers show the gap between the two.
| What the marketers said | Share |
|---|---|
| Their organization uses AI-powered applications | 95 percent |
| AI made them more productive | 87 percent |
| AI made their content perform better | 39 percent |
The work got faster for most of them, and far fewer saw their content perform better. My reading of that gap: every task still waits for a person to start it, carry it through and finish it.
As I see it, the ways people use AI fall into three roles, and each one builds on the one before.
| Role | What the person does | What it looks like |
|---|---|---|
| Search box | Asks a question, takes the answer, does the rest by hand | A question in, an answer out |
| Sparring partner | Thinks with it, tests an argument, drafts with it, and keeps every step in their own hands | Useful, and still assistance, because the person carries the work from one step to the next |
| Executor | Enters the goal and judges what comes back | One campaign goal in, the whole campaign back for review |
Most people see AI as a chat window. In the organization I work inside, with 62 AI seats, 4 people used one daily and 48 used one at least monthly. About 25 of those 62 seats belonged to product rather than to sales or marketing. Most users averaged about one message per session, a question in and an answer out. I wrote that case up in why nobody uses the AI seats you bought.
Agents are one sign of the executor step, and they are still rare. Gartner's 2026 survey of CIOs and technology executives found that 17 percent of organizations have deployed AI agents, while more than 60 percent expect to within two years.
What does a normal week look like in an AI-first company?
In an AI-first marketing team the week starts with the vision, and the system turns it into a first draft. That is how I expect it to run, and parts of it already run in the organization I work inside.
On Monday the team meets and sets the vision for the next two or four weeks:
- The team agrees on what it wants to reach, using what the data from its tools says.
- The marketer puts that vision into the system.
- The system uses the connected data, from web analytics, the CRM and the social channels, and turns the vision into a first draft that is close to ready.
- The people judge the draft, and either send it back with what needs to change or release it.
What the person brings decides the result. The more detailed and the more creative the input, the better the result.
For a rep, the morning starts with a briefing on the accounts assigned since the day before, with contacts for up to three key roles already researched. That routine runs on workdays at 07:30. What I expect next: replies drafted from the company's own knowledge, so that what is left is sending them and picking up the phone.
Who talks to whom changes as well. People work with each other more, as a team, to get it done, and they work with the system more, to turn it into a result. The question people ask is how the team's vision gets done.
What happens to the prompt in an AI-first company?
The prompt changes shape: the method that used to be typed into it moves into the system. I do not think I ever really stopped prompting. What changed is what goes into the prompt.
Before, the best practice was a prompt with six parts:
- a role, such as "you are an experienced performance marketing manager"
- a goal
- the task
- the steps along the way
- what it must never do
- what the output should look like
The person now types one simple prompt in plain language, and agents in the system break it apart into those parts. So the prompt says what we want to do, hands over the knowledge and the briefing, names the goal, and then: go. What that buys is reuse, and a prompt quality that stays the same.
Where the instructions live matters the moment more than one person uses them. I built a daily account briefing meant for every rep. With the instructions in each rep's prompt, that would have meant 20 versions. In a skill, a written instruction file the system reads, it meant one file and one sentence to start it.
After the result comes back, the person judges it in plain words and says what worked, and that gets written down for the next task to start from. That step is where most systems fall short. MIT's Project NANDA found in 2025 that the barrier to scaling AI is learning: most systems do not retain feedback, adapt to context or improve over time.
What has to exist before anything runs on its own?
The company's knowledge has to exist in writing first, because the system can only follow what it can read. When a company decides to work this way, the first thing missing is the description of the company itself:
- the positioning, the offering and the messaging
- how the company presents itself, its brand and its colors
- what it knows about its existing customers
- who it sells to: its ICP, meaning the profile of the companies it wants as customers, and its personas
Without that there is nothing for the system to hold itself to. A company that has documented none of it will find it very hard to set this up on the day it decides to, so it is not ready yet, and its first step is writing it down.
The second requirement is where the work comes from. What gets automated comes out of the processes a company already runs every day: the campaign, the account research, the follow-up. It comes out of daily work, never out of a wish list.
How do you know the work actually runs on its own?
Ask when it last ran. Being built and running are two different facts.
Starting without a reminder is the condition that fails quietly. Inside the B2B GTM organization I work in, the follow-up sequence for a campaign was built and configured, and from 3 August on it did not send a single mail. Seven weeks passed with no error and no alert, and nobody noticed, because nothing was broken. I wrote that case up in the follow-up sequence that sat silent for seven weeks.
The date of the last run moves every time the process runs, so an old date, or no date at all, is the sign that it is not running. A routine of mine showed the same thing when it failed quietly and went on looking healthy.
What happens to a company that does not run on AI?
In my view its competitors pull ahead. Something will still work without AI. The question is whether a company can keep up, while its competitors will be faster, more effective and more efficient. And over time they will take its customers.
It is about output, and it is never about cutting people. The same team can deliver much more, and possibly better, because nobody has to be hired anymore for know-how that is more or less common knowledge. Building a LinkedIn or Meta ad campaign used to mean hiring somebody who knew how. Today anybody can find out how to build one. Whether the campaign performs is a different question, and that is where the people come in.
The tools are moving the same way. Instead of clicking through HubSpot to find out what is in the CRM, I can ask the AI connected to it a question in plain language. Instead of learning what Clay can and cannot do, I can describe the process, have it built through the command line, test it once, and if it runs, it runs. Many tools are going headless: the interface a person had to learn matters less and less.
Why I think every company has to make this move is a longer argument, and I wrote it down in why every company will have to run on AI.
How do you tell whether your company is AI-first?
Ask three questions, one for each condition, and then one more:
- Which recurring task starts without anybody remembering to start it? If you cannot name one, your company uses AI and does not run on it yet.
- When it runs, does anybody nudge it from one step to the next? If somebody does, it is assistance with a trigger in front of it.
- Does a person judge a finished result and release it, or approve every step along the way? If every step needs a yes, the person is still carrying the work.
- When did the task from the first question last run? The answer is a date. If nobody can give one, nobody is checking.
Find out whether your company is ready for a first build.
We check three things with you first: whether leadership is behind it, whether you have decided which AI system to commit to, and what a first build can and cannot do. Then you get a yes or a reasoned no.
Build your AI-first organization