The Agency Model Is About to Get Restructured. Most Founders Won't See It Coming.

This is part one of a six-part series on rebuilding the agency model for the autopilot era. We work with founder-led agencies at $1M-$10M on exactly this transition at Agency Focus. 

A few weeks ago, Sequoia partner Julien Bek published a thesis that will quietly reshape the next decade for services businesses. You can read the original here. It is worth your forty-five minutes.

The headline claim is that the next trillion-dollar company will not sell software. It will sell the work.

Not a tool for accountants. The firm that closes the books.

Not a tool for lawyers. The firm that reviews the contracts.

Not a copilot for insurance teams. The company that handles the claims.

Most of the commentary on this piece has focused on the math. For every dollar a company spends on software, it spends roughly six dollars on services. In accounting, the ratio approaches 1:12. Every SaaS founder chasing the software budget has been fighting over the smaller pile the whole time. That math is real, and it matters.

But the math is the reason why. Agency founders need to internalize the part of Bek's thesis that explains how to think about their own work. And that sits in two frames that most people skimmed past.

Bek's core distinction is between intelligence work and judgment work.

Intelligence is rule-based pattern work. The rules can be complex, but they are still rules. Translating a spec into code. Examples: Coding medical claims into ICD-10. Drafting an NDA from a standard template. Shopping insurance carriers to find the best rate. Screening resumes against a defined criterion. The work requires training and attention, but not taste.

Judgment is experience-based taste. It is knowing which feature to build next, not how to build it. It is reading a room and knowing when to push and when to back off. It is the ninth proposal you wrote that taught you what the second one got wrong. Judgment is built over years of work and cannot be documented.

AI has crossed the threshold at which it can perform most intelligent work autonomously. Judgment is still firmly human. The higher the intelligence ratio in any given category, the faster that category gets restructured.

Software engineering was the first profession to hit the threshold. A year ago, most developers were using AI as an autocomplete. Today, agents are starting more tasks than humans. Software engineering now accounts for more than half of all AI tool usage across every profession combined. Every other category is still in single digits. Not because the other categories are safe. Because software engineering is primarily intellectual work, it got there first.

It is coming for every profession with a meaningful intelligence component. Which is almost all of them. Which is almost all of us.

Here is the part agency founders need to sit with for a minute.

Look at your last five engagements. Be specific. Map every deliverable onto the intelligence-judgment spectrum. Campaign reporting. Research decks. Competitor audits. Landing pages. Email sequences. SEO technical fixes. Brief writing. Most of what you charged for was intelligence work. Pattern recognition and rule application, and template plus input equals output. A well-designed system with a smart operator can now do that work at a fraction of your cost.

The judgment portion, the genuinely hard stuff, is smaller than you want to admit. Choosing which market to target first. Knowing when to kill a campaign. Reading a founder and telling them the truth they do not want to hear. Pricing a deal to land it without giving away margin. That is real judgment. It is defensible. But in most agency engagements, it’s about twenty percent of what you deliver and eighty percent of what you should sell.

The agencies that survive the next cycle are the ones that figure out which of their work is intelligence and which is judgment, and then restructure everything around that split.

Bek's second frame is the delivery model itself.

A copilot puts AI in the hands of a professional. The professional still does the work. The client still pays the professional. Harvey sells to law firms so lawyers can draft faster. Rogo sells to investment banks so analysts can model faster. The customer is the professional. The output is still the professional's output. The tool just makes them go faster.

An autopilot skips the professional entirely. Crosby ships the NDA directly to the company that needs one. WithCoverage closes the insurance policy for the CFO without a broker in the loop. ColdIQ books the sales meetings for the B2B company without the SDR. The customer is the buyer who actually wants the outcome. The tool is invisible underneath. The agency, the broker, the firm are not in the transaction.

Most agencies right now, if they are using AI at all, are accidentally becoming copilot shops. You bolt AI onto your existing delivery. Your team is faster. Your margins tick up a little. You charge the same rates. You feel like you are winning.

You are not winning. You are running in place. Your competitor two years from now is not another agency with better AI tools. Your competitor is an autopilot company that sold the same outcome directly to your client and doesn't need you in the middle.

This is the innovator's dilemma that Bek names at the end of his piece. The copilot company that tries to become an autopilot has to cut its customers out of the work. The agency that tries to pivot has to stop selling to the person who writes the check today and start selling to the person one level up, who just wants the outcome. That is a harder sale. It is also the only sale that still exists in five years.

Take demand generation as the working example, since that is the corner of the world I spend most of my time in.

A copilot demand gen agency in 2026 uses AI to write emails faster, build lists faster, and analyze campaigns faster. They sell the same $ 25K-per-month retainer they sold in 2023. The deliverable is activity, measured in emails sent, calls booked, and reports delivered.

An autopilot demand-gen company sells 40 sales-qualified meetings per quarter in the enterprise segment, priced per meeting above a floor, and reports live. The client never sees the system underneath. They do not need to. The AI-native infrastructure handles list building, personalization, sequencing, reply handling, and scheduling at a cost that the Copilot shop cannot match. The autopilot captures more of the outcome-based budget that previously went to an in-house SDR and a sales tool stack.

Same budget line on the client side. Entirely different on the agency side. Different margins. Different defensibility. Different relationship.

Vacca's ColdIQ crossed seven million ARR running exactly this model. Crosby is shipping legal work. WithCoverage is closing insurance. The list is getting longer every quarter, and the categories where it works keep expanding as the models improve.

The two paths for agency founders are cleaner than most people want to admit.

One: You keep selling intelligence work as if it were judgment work. You add AI tools to deliver faster, and you pass most of the savings to clients because everyone else is doing the same thing. You compete on relationships, creative chemistry, and pricing pressure. You add heads to grow. Your margins compress. The smart offshore shops and the AI-native autopilots squeeze you from both sides. This is the Copilot-era agency on the way down.

Two: you separate your intelligence work from your judgment work, rebuild the intelligence layer as a system, and start selling outcomes to buyers who want the work done, not done faster. Your team stays small. Your margins widen as the models improve. The data you accumulate about what works in your vertical becomes a moat that no new competitor can spin up. You become an operating system for a specific outcome in a specific category. This is the autopilot-era agency on the way up.

Neither path is easy. The first one is death by a thousand cuts over three to five years. The second one is a deliberate rebuild that takes 18 to 36 months and breaks a lot of things along the way.

The reason to start now is not that the technology is finished. It is not. The reason is that the agencies that start rebuilding now will have the documented systems, the outcome-priced client book, and the compounding data asset in place by the time the wave actually breaks. Those who wait until their revenue is shrinking will be restructuring under pressure, which is the worst possible time to rewire a business.

I am writing this series because it is the conversation I have with every agency founder I work with right now. Some of them are excited. Most of them are quietly terrified. A few are pretending it does not apply to them, which is the response I am most worried about.

Over the next five posts, I am going to walk through what the autopilot-era agency actually looks like. How pricing changes when you sell outcomes instead of activity. How the delivery stack changes and the hiring order that matters. What data becomes the moat, and how to start compounding it on Monday. How to sell the new model to clients who were trained to buy the old one. And finally, what do the first 90 days of the transition look like for an agency with $2–5 million in revenue that wants to start without breaking the business?

If you are running an agency right now, the question I would ask you before next week is this one.

Map your last engagement on Bek's two axes. How much of what you delivered was intelligence work that a well-designed system could do? How much was genuine judgment that only a human with your experience could hold? And then the harder question: how much of what you charged for was priced as judgment when it was actually intelligence?

The honest answer to that last question is the starting point for everything that comes next.

Next week: Stop selling hours and start selling outcomes. What actually changes when you do this?