In the Future, There Will Be Only Three Kinds of Companies

A while ago, I was scrolling through the article list of a marketing peer of mine.

Everywhere I looked, the same word.

AI.

Not the vague "AI is coming" kind of talk. The very specific kind — competitive analysis in one minute, an AI Agent writing your cold-start outreach scripts for you, letting it chew through that messy pile of sales data.

I read them one by one, and by the end, only one sentence was left in my head:

In the future, there will be only three kinds of companies.

Which three?

AI-native, AI-emergent, and obsolete.

What does AI-native mean? It means the company, from day one, was designed around the idea that "every desk has an AI assistant sitting at it." The workflows, the roles, even the hiring bar — all rewritten to fit AI.

What does AI-emergent mean? An old company, but one where everyone is scrambling to wedge AI into their own work. Maybe still clumsily, maybe only halfway — but at least moving.

And obsolete?

Obsolete is when you're still doing today's work the way you did it ten years ago.

That line isn't mine. But the more I think about it, the truer it feels.

But what I want to talk about today isn't these big words

Anyone can say the big words. AI-native, AI-emergent — you nod along, go home, and do whatever you were doing before.

What's actually interesting are the things specific enough to copy.

In that peer's articles, I came across a few things that made me go "oh?" out loud.

The first: competitive analysis in one minute

How long did competitive analysis used to take?

The slowest I've ever seen: two weeks. Pull a team, pick a framework, scrape data, meet, scrape data again, meet again. Finally a PPT lands — and the competitor has already changed strategy.

And now?

Under a minute.

You dump the competitor's website, product pages, pricing, and user reviews into AI all at once, and tell it what you want to know. What it hands back isn't a pretty report — it's a few real questions: on which customer segment is the competitor stronger than you? What concession is hidden in their pricing structure? Which selling point did you miss?

Wow.

This isn't "saving time." This is compressing a two-week job into the time it takes to take a sip of water.

But let me remind you of one thing — the quality of AI's answer depends on the quality of your question. Same one minute: some people ask it for nonsense, some people ask it and strike gold.

The second: let AI chew through that pile of messy data

What's the biggest headache for marketers?

Not having no data. Having too much data, and it's a mess.

A pile in the CRM, a pile in the ad dashboards, another pile in the customer-service systems. The fields don't line up, the definitions don't line up, the timestamps don't line up either. You want to see which channel brought the highest-quality leads last month — just merging the data into one table takes half a day.

Now there are tools like OpenAI's Codex and Anthropic's Claude Code — I know, the names sound like something for programmers (they're coding- and agent-style tools).

But don't let the names scare you off.

At its core, it's like hiring an intern who never sleeps, never complains, and can read ten spreadsheets at once.

You tell it: "Match the last three months of ad spend with the deals that actually closed in the CRM, and tell me which channel has the highest actual conversion rate — not the highest click-through rate (CTR)."

And it does it.

Think about it — this kind of work used to mean either grinding through it yourself, or paying someone to do it. Now it's one chat window.

The third: an AI Agent runs your cold-start for you

What's an AI Agent?

Let me give you the simplest way to understand it — an Agent is AI that does the work itself, not just answers your questions.

You ask ChatGPT, "How do I write a cold-start email?" (cold-start = outreach to prospects you have no prior relationship with) It hands you an article.

But you give an Agent a task: "Write a personalized opening email for each of these 50 prospects; mention their company's recent news; keep the tone not too formal." It actually goes and looks things up, writes them, lines them up.

When it's done, you glance through them, and send the ones that work.

Strong stuff.

But there's one line in that peer's article I want to underline — an Agent can do a lot of things, so the key is that you first figure out which one thing you want it to do. It's not that it can't — it's that you haven't thought it through.

And then, I have to be a bit of a buzzkill

Now that we're done being excited, let's talk about the risks.

Otherwise this isn't real.

Risk one: the AI you depend on today might be gone tomorrow

A lot of people haven't realized this.

AI models get replaced incredibly fast. Today you lock your whole workflow to one model — tomorrow it shuts down, changes its API, raises prices, or just stops being good — and your work is broken.

So what do you do?

The peer gives a plain but useful piece of advice: write down the things you rely on AI to do, and crystallize them into your own methodology. Models will leave; your method can't leave. Prompts, processes, judgment criteria — those are your assets, not some model's.

Risk two: the AI bill burns more money than you think

This one I've felt personally.

AI is so smooth to use that you don't notice you're burning money. Try this tool today, subscribe to that service tomorrow, the day after, upgrade the whole team to the premium tier — and at the end of the month you look at the bill, and it's higher than half a year of the marketing department's budget.

How do you break the cycle?

Try less, use more. Go deep on the one or two scenarios you've already validated, instead of chasing new toys all over the world. Experimentation costs money, and it usually costs more than you think.

Risk three: there's an AI divide inside your team

This one's the most hidden.

In the same team, some people are already flying with AI, and some are still stuck at "I heard it's really powerful." At first the gap is small — six months later, the people who are fluent are producing three times, five times as much as the people who aren't.

And then comes the rift.

Once a rift opens, it's incredibly hard to close. Because the fluent ones get more fluent, and the hesitant ones get more anxious, more afraid to even touch it.

So you need a system. Make AI learning inside the team continuous and shared, not everyone practicing on their own.

Finally, I want to say something even more valuable

All of that — competitive analysis in a minute, an Agent writing emails, chewing through messy data — is still treating AI as a tool.

But there's a line in that peer's article that made me stop and think for a long time.

Treat AI as a thinking partner, not just a content machine.

What does that mean?

The vast majority of marketers use AI to write copy, write emails, write tweets. Honestly, they use it as a typing pool.

But where AI is truly valuable is when you throw it a question you haven't quite figured out, and let it reason through it with you.

"Is there a hole in my strategy this quarter?" "If the competitor counters like this, how do I respond?" "Is this judgment of mine actually just because I got yelled at by the boss last week, and emotion is taking over?"

It can give you a calm rebuttal that your emotions won't drag off course.

Honestly. That value is far greater than having it write ten thousand tweets.

To wrap up

Back to that opening line.

In the future, there will be only three kinds of companies.

But don't let that line scare you. What decides which one you are isn't how big the company is or how much money it has — it's whether you and your team, today, actually get your hands dirty, even clumsily, and wedge AI into even one small thing.

It doesn't have to be perfect. Wedge it in first.

That's the thing I most wanted to say to you after reading that pile of articles.

Here's to you — getting your hands dirty.