Can an AI Agency Actually Exit for $100M?

There's a version of the AI agency story that ends with a lifestyle business—a few clients, comfortable income, no boss. Devin Kearns, co-founder and CEO of Custom AI Studio, wants to tell you about the other version. The one with a $100 million exit target, a deliberate march up-market, and a fairly unsentimental view of what most people currently selling "AI solutions" are actually building.

In a nearly two-hour conversation with YouTuber Nate Herk, Kearns laid out the strategic framework behind Custom AI Studio's evolution—from $2,500 n8n automations in the early days to $400,000–$500,000 annual contracts with mid-market companies. It's a compelling story, but the more interesting thing is the argument underneath it: that the AI agency space is about to sort itself into winners and people who didn't notice the game changing.

The Commodity Problem Nobody Wants to Say Out Loud

Kearns opens with something most builders in this space would rather not hear directly: "The value of actually doing development is trending towards zero. It's like very very stark and very very clear that it's happening."

His evidence isn't abstract. He describes a 67-year-old lawyer who showed up with a vibe-coded app that was, frankly, decent. When the technical barrier to shipping software drops that far, charging premium hourly rates for development becomes a harder and harder sell. This isn't a distant-future problem—it's the ground shifting right now, under active practitioners.

The implication is worth sitting with. A lot of the AI agency market that emerged over the last two years was built on a skill arbitrage: builders who understood n8n, LangChain, or agent architecture had capabilities that their clients didn't, and they could charge for that gap. That gap is narrowing fast. The tools Kearns and Herk both mention having built extensively—RAG pipelines, n8n workflows—are things neither of them would reach for today. Claude and its successors have made entire tool categories redundant in a matter of months.

What this means for the people currently selling those tools as services is the uncomfortable center of Kearns' argument.

Why Mid-Market, and Why It's Not Obvious

Kearns' bet is on mid-market companies—roughly $10M to $250M in annual revenue—as the sweet spot for AI consulting engagements. His reasoning is structural rather than size-based: these companies have already had to build repeatable systems. They've written SOPs. They've hired teams and documented processes to manage them.

"The pattern across 97% of the projects that we've done is when you really break it down fundamentally what we are doing is we're taking the logic of the business—whether they've written it down or not—and we're just converting it into an AI system."

That matters because converting existing logic is a very different problem from inventing it. With an SMB or solo founder, the "logic" often lives inside one person's head and hasn't been systematized. You're not just building an AI system; you're also doing the organizational design work of figuring out what the process should even be—a scope that tends to expand indefinitely.

With enterprises, ironically, Kearns argues the opposite problem: they're often less systematized than mid-markets because they've always had enough bodies to throw at problems. They can survive with inconsistency in a way a 50-person company cannot.

The mid-market sweet spot, by this logic, isn't about budget or prestige. It's about having enough process infrastructure to give AI something coherent to automate—and a clear enough P&L impact that success can actually be measured.

The e-commerce example he walks through is instructive. A company processing 40,000 support tickets a month had a 21% refund rate on their main product. They'd already mapped out the refund-pushback logic, built decision trees, written the SOPs—they just couldn't get their human team to execute consistently. Custom AI Studio deployed an AI system against that existing logic and brought the refund rate from 21% down to 16%. The client had asked for 1–2% improvement. They got 4–5%, unlocking millions in working capital and resetting their entire growth trajectory.

This is the kind of project that justifies moving up-market. A comparable percentage improvement for a $500K revenue company is a nice win. For a company doing tens of millions, it's transformational—and that transformational value is what enables the pricing that builds real enterprise value.

The Exit Math Most Practitioners Skip

Here's where Kearns gets into territory most people in the AI agency space genuinely haven't thought through. He maps out an enterprise value framework: at $2M annual revenue as a pure AI readiness consulting firm, your business is essentially worth its annual revenue—maybe $2M. But get past the $5–6M mark and the acquisition multiple jumps dramatically, from roughly 1–2x EBITDA to something closer to 5x. The same business that was worth $2M at $2M revenue becomes worth $30M at $6M revenue, not because anything structural changed, but because you've cleared the threshold where institutional buyers see a scalable asset rather than a founder-dependent practice.

Kearns is explicit that this is borrowed from Hormozi's acquisition framework, and that the specific numbers are rough guidelines rather than guarantees. But the underlying logic—that services firms rerate at scale, not by category—is real and documented in the M&A market.

The reason this matters isn't just for people who want a big exit. It's that the exit target determines the build strategy. If you're aiming to sell a lifestyle business, you optimize for personal income and low overhead. If you're aiming for a legitimate acquisition, you need recurring revenue, documented processes, team capacity that isn't solely dependent on the founder, and a client base that a buyer could actually inherit. Those are different companies, and you can't really work backwards from one to get to the other.

What They're Actually Selling

One of the more useful reframes in Kearns' framework is about what mid-market clients are actually buying. He describes the internal pressure these companies face: boards demanding AI strategies, LinkedIn full of competitor announcements, internal teams split between AI enthusiasts and people who genuinely hate the technology. "They're not buying an AI system or an automation. They're buying relief."

The relief is being able to say, credibly, that you have an AI strategy—one that's been validated by an outside expert, is grounded in your actual operations, and can be explained to your board without embarrassment.

This reframe has practical implications for how you sell. Kearns admits he made the mistake of drifting toward technical language on sales calls as his own expertise deepened. The correction was recognizing that the client's problem isn't a technical one—it's an anxiety-and-credibility one. The AI system is how you solve it; it's not the thing you lead with.

His sales process now follows a sequenced trust-building model: an initial workshop to align on the current AI landscape, a discovery phase (what he calls the "blueprint," priced at $15K–$35K) that maps business logic and identifies high-ROI opportunities, then a custom build, and eventually—for the right clients—an ongoing "AI technology partnership" structured more like a growth-marketing revenue share than a time-and-materials contract.

The Questions Kearns Doesn't Fully Answer

There are real tensions in this framework that the conversation surfaces without resolving.

The rev-share partnership model Kearns describes—where Custom AI Studio takes a percentage of revenue improvement rather than flat project fees—is compelling in theory and genuinely hard in practice. Attribution is straightforward when you improve a refund rate. It's murky when you're improving underwriting processes for a real estate investment firm. Even Kearns acknowledges the accounting example he's working on doesn't have a clean P&L through-line. The model requires the right client, the right KPI, and enough trust that both parties agree on what "caused" the outcome. That's a high bar.

There's also the implied question of timing. Kearns is arguing that now is the inflection point—that the early majority is finally coming online to enterprise AI, which creates a window for agencies that got in early to establish themselves as trusted partners before the space gets crowded with McKinsey and Anthropic's own consulting arm (the $1.5 billion Blackstone/Goldman enterprise services firm they discuss is very much pointed at this market). That window thesis might be right. It might also be right and narrower than expected.

And then there's the honest caveat Kearns himself drops: "Honestly, most people probably shouldn't do their own thing. It's probably not ideal because it's competitive and most people fail."

That's not a dismissal of the opportunity. It's a reminder that the gap between "AI is a real business opportunity" and "I should start an AI agency" runs through a lot of territory that includes sales tolerance, operational stamina, and genuine domain expertise—not just the ability to build something technically interesting.

The $100 million exit is a real number with a real path. Whether it's your number is a different question entirely.

---

Marcus Chen-Ramirez is a senior technology correspondent at Buzzrag. He spent eight years as a software engineer before he realized he was better at explaining things than building them.