The AI company did not replace the agency. It bought one. That decision says more about what software cannot do than any product announcement, and it should change how you think about buying growth.

Key takeaways

• When an AI agent company acquires the agencies it set out to automate, it is admitting that client relationships and judgment are not features you can ship.

• Automating the agency is not a product. It is an operating model, and operating models need people who know the client.

• The right question for any growth partner is no longer "what will you do for us" but "what will you take off our plate so we can be with our customers."

The Acquisition That Admits Something

Two stories crossed the wire within days of each other last week, and together they say something neither says alone.

The first came out of Tel Aviv. Enso, a startup that builds AI agents for marketing and sales, announced plans to acquire two marketing firms for twenty to twenty-five million dollars: a performance agency managing roughly two hundred million dollars in media, and a go-to-market firm in the United States. Enso has twenty-six employees. The deals would roughly triple its headcount, most of it in exactly the kind of agency work its agents were built to automate.

The second came from Advertising Week in New York, where the media chief of Liquid Death described rebuilding the brand's marketing around AI with its agency, Power Digital. The ambition is to run a sizable budget with two people inside. Campaigns once optimized weekly now get optimized daily. He said he never wants to see a spreadsheet again. The agency, in this arrangement, is no longer extra hands. It is the system the two people stand on.

Read together, they are a confession. The company that set out to automate agencies decided it needed to own some. The brand that wants the smallest possible team decided it needed a partner to build its infrastructure. Both reached the same conclusion from opposite directions: the software was not enough.

Operating Model, Not Product

There is a simple reason the software was not enough.

An AI agent can test a hundred landing pages overnight. It cannot decide that the client's real problem is a pricing page that embarrasses the sales team. It can draft an outreach sequence in seconds. It cannot know that a founder's reputation in a small market depends on never sending a cold email that sounds like one. It can optimize a media budget daily. It cannot sit in the room when the budget gets cut and help the owner decide what to protect.

Those are not edge cases. They are the work. Everything else was always going to be automated, and most of it should be. The mistake is to believe that once you automate the repetition, the judgment comes free. It does not. Judgment lives in people who know the client, the market, and the moment. Enso found that out and bought fifty of them.

This is why "automate the agency" has never been a product. It is an operating model: a way of pairing people who own judgment with systems that own repetition, inside a relationship with a specific business. You cannot download an operating model. You build it around the client.

We learned this the slow way at Stellant, because we started from the other end. We began with senior people and added the systems. Enso began with systems and is now adding senior people. The shape you end up with is the same, and the shape is the point.

The Two-Person Marketing Department

The Liquid Death story is the more useful one for a small or mid-sized business, because the number in it is the number you will be asked about.

Two people. A large budget. Daily optimization. The instinct is to ask whether you could run your marketing with two people. The better question is what those two people would spend their day doing.

In the version that works, the two people are not operators. They are not pulling reports or moving budget between campaigns. The system does that. They are deciding what the brand is willing to say, listening to customers, holding the agency and the AI to outcomes that matter, and staying close to the rest of the company so marketing stays connected to what is actually being sold.

In the version that fails, the two people are the same operators as before with more dashboards. Headcount went down and the work did not change. They are busier, more isolated, and further from customers than the five people they replaced.

The difference is not the technology. It is whether the leaders decided what people are for.

The Judgment Ledger

Here is a way to make that decision for your own business this week. It takes a sheet of paper and a quiet hour.

Draw two columns. On the left, list every recurring piece of growth work your business does: content, email, ads, reporting, website changes, lead follow-up, proposals, reviews, events, partner outreach. Most owners find twenty to thirty items.

Now ask one question of each: if this were done perfectly but mechanically, with no understanding of our customers or market, would it still be worth doing? If yes, it moves to the right column under repetition. If no, it stays on the left under judgment.

Most owners find the list splits roughly seventy-thirty. Seventy percent of growth work is repetition worth doing well that does not need a senior person. Thirty percent is judgment, and it is where almost all the value comes from.

Then look at your calendar and payroll. If your most experienced people are spending their time in the repetition column, you have found your problem, and it is an allocation problem, not a technology problem.

Finally, write the name of the person who owns each judgment item. If the same name appears on all of them, you have found the bottleneck, and it is probably you.

From Insight to Action

1. Run the Judgment Ledger before your next growth conversation, so you walk in knowing which thirty percent you are protecting.

2. Move your most experienced person out of the repetition column this month, even for one recurring task. Watch what they do with the time.

3. Ask every partner which of your tasks they will own outright, not assist with. "We help with content" is a different promise than "you will never think about content again."

4. Decide what your two people are for before deciding whether you can get down to two. Headcount follows the decision, not the other way around.

5. Put a customer conversation on the calendar for each hour the systems give back this quarter. Make it a rule, not an intention.

6. Watch who is buying whom. When AI companies start buying service firms, the service firms were never the problem. The question was always who owned the judgment.

A company that automates the repetition defaults to the same work done faster. A company that protects the judgment defaults to growth.

Sources

CTech by Calcalist: "Enso to acquire two marketing firms for up to $25 million to bring AI agents into advertising," October 5, 2026

Marketing Dive: "How Liquid Death is building an AI-native marketing organization," October 7, 2026