AI Is the New Agency

I’ve been thinking a lot about a simple (and slightly provocative) idea:

AI companies are the agency companies of a new era.

Not because they sell “services” in the traditional sense, but because AI changes the shape of cost curves—so the old “SaaS beats agency” story starts to wobble.

This is a personal reflection, not a universal law. But the pattern keeps showing up across products, pricing pages, and investor conversations, so I want to put the thought in writing.

The Old Trade: Why SaaS Got the Multiple

In the pre-AI era, the contrast between agencies and SaaS was clean:

• Agency businesses sold outcomes with human labor. Great agencies created enormous value, but the economics were stubborn:
  • Labor was expensive and hard to scale.
  • Marginal cost stayed high: more clients usually meant more headcount.
  • Quality often depended on specific people.
  • Revenue was less “purely recurring” and harder to compound.
• SaaS businesses sold software. The dream was simple:
  • Build once, sell many.
  • Marginal cost per additional user was close to zero.
  • Gross margins were high and predictable.
  • Recurring revenue and expansion could compound.

That’s why markets historically rewarded SaaS with higher valuation multiples. The underlying assumption was that software turned labor into leverage, and leverage into margins.

AI Changes the Cost Curve (Especially for Agencies)

Now enter AI.

AI doesn’t eliminate human work, but it re-prices it. A capable operator with the right tools can do the work of a small team: drafts, research, QA checklists, first-pass designs, code scaffolding, content variations, internal documentation, analysis. Not perfect—but fast, and often “good enough” to reach a human review stage.

That does two things for agency-style businesses:

1. Effective labor gets cheaper. You still pay for taste, judgment, and client context—but the “blank page” cost drops.
2. Throughput per person increases. One team can serve more clients without quality collapsing immediately.

Historically, the agency ceiling was headcount. AI raises that ceiling.

In other words: the thing that made agencies look “unscalable” is being attacked directly.

When Software Stops Being Near-Zero Marginal Cost

At the same time, AI introduces a new kind of cost into software:

• Tokens
• Inference time
• Context windows
• Tool calls
• Retrieval
• Fine-tuning / evals
• Human-in-the-loop for edge cases

In a traditional SaaS app, once you’ve built a feature, another user mostly adds negligible cost. In an AI-powered product, another user can add real variable cost—sometimes meaningfully so.

This doesn’t mean “SaaS is dead.” It means some AI-first SaaS starts to resemble something we used to associate with services:

• A cost that grows with usage
• Capacity planning (not just servers, but model usage)
• Margin sensitivity to how customers behave

If you sell “unlimited AI” for a flat subscription, you’re taking on an agency-like risk profile: unpredictable workload with real marginal costs.

So the old mental model flips:

• Agencies get leverage from AI and feel more scalable.
• SaaS absorbs AI compute costs and can feel less “infinite margin” than before.

AI Companies as Outcome Machines

This is where I think the “new agency” framing becomes useful.

Many of the most compelling AI companies don’t sell software seats; they sell a job getting done:

• “Turn these leads into booked meetings.”
• “Process these invoices and reconcile them.”
• “Keep our knowledge base updated and answer support tickets.”
• “Draft, personalize, and sequence outbound.”
• “Generate creative variations and learn what converts.”

This is agency language: outcomes, deliverables, responsibility.

But the mechanism is different:

• Instead of scaling by hiring more people, you scale by improving workflows, prompts, tooling, evals, and orchestration.
• Instead of “billable hours,” you optimize “cost per completed task.”
• Instead of pure software adoption, you often need onboarding, process change, and integration—things agencies were always good at.

An AI company that promises an outcome ends up doing a lot of what agencies do:

• Understand messy constraints
• Work across tools
• Deal with exceptions
• Own the last mile

The product isn’t just an interface. The product is operations with software inside.

What This Means for Pricing (and Why It’s Hard)

If your costs are variable, pricing becomes strategy, not just packaging.

I see three broad pricing paths:

1. Seat-based (classic SaaS): easy to understand, but can break when power users generate huge compute bills.
2. Usage-based: aligns cost and revenue better, but customers hate surprise bills and will game the meter.
3. Outcome-based: closest to the “agency” spirit—charge for results, not tokens—but requires tight instrumentation and clear definitions of success.

My guess is we’ll see more hybrid models: base subscriptions with usage tiers, plus “done-for-you” add-ons that are basically productized services.

And importantly: compute costs will become a first-class part of product design. Caching, routing, model choice, truncation, and evaluation aren’t just engineering details—they’re margin decisions.

A Market-Level Prediction: Multiples Will Follow Unit Economics, Not Labels

If this framing is right, then the classic “software multiple vs services multiple” gap narrows—not because investors stop caring about margins, but because the category boundaries blur.

In the AI era, the question becomes less “Is it SaaS?” and more:

• What is the contribution margin per task?
• How fast does it improve as models get cheaper and workflows get better?
• Can you retain customers when you’re selling an outcome (not a tool)?
• Do you own distribution, or are you a thin layer on top of someone else’s model?
• Are you building a system that compounds (data + feedback + workflow), or a one-off integration shop?

The best businesses will still earn great multiples—but they’ll earn them by proving durable economics, not by wearing the “SaaS” label.

My Bet: The Winners Look Like Agencies First, Then Like Platforms

Here’s the pattern I expect to see repeat:

1. Start agency-like. Go deep into a workflow. Own the outcome. Do the integrations. Learn the edge cases.
2. Productize the playbook. Turn repeated operations into software. Build evals, monitoring, and guardrails.
3. Scale with leverage. As the workflow becomes more reliable, you need fewer humans per customer.
4. Earn platform dynamics. If you become the system of record for an outcome, you get retention, expansion, and ecosystem pull.

This looks like an agency evolving into a product company—but with AI making the transition faster and cheaper than before.

Closing Thought

For years, “SaaS” meant low marginal cost and agencies meant high marginal cost.

AI scrambles that mapping.

Agencies gain leverage. Software inherits variable costs. And the companies that win may be the ones that stop arguing about what category they’re in—and instead obsess over the economics of one thing:

How cheaply and reliably can we deliver a valuable outcome?