The famous “$1 software, $6 services” idea has been floating around Silicon Valley for years. The premise is simple: for every dollar spent on software, companies spend roughly six dollars implementing, operating, customizing, integrating, managing, and extracting value from that software through services.

What is interesting now is that AI companies are all implicitly claiming they can capture the entire $7.

Not just the software dollar.
All of it.

The software.
The implementation.
The execution.
The operations.
The optimization.
The services layer.
Everything.

And honestly, I think a lot of founders believe this is inevitable because AI dramatically compresses labor costs. If a system can generate code, generate creative, generate campaigns, generate reports, generate analysis, then the assumption becomes: “obviously the software company now captures the services revenue too.”

But I think most people are massively underestimating how difficult that transition actually is.

Because replacing labor is not the same thing as owning outcomes.

Most agencies right now are making the exact same mistake. They are using AI primarily to reduce labor cost inside the existing agency model. Faster reports. Faster creative generation. Faster SEO content. Faster media analysis. Faster onboarding. Faster decks. Faster summaries.

That is not a new operating model. That is the old operating model with better tooling.

You still fundamentally have the same fragile structure underneath:
humans moving information between systems,
humans manually coordinating workflows,
humans stitching together context,
humans translating between strategy and execution,
humans constantly re-explaining the same things to different people.

The result is that many agencies are becoming “cheaper execution engines” instead of becoming true growth infrastructure companies.

And I think that is the wrong game entirely.

The real opportunity is not selling cheaper labor. The real opportunity is selling managed growth loops.

That distinction matters because agencies historically sold hours, then deliverables, then expertise. But none of those things are actually what the customer wanted. The customer wanted movement. Specifically, movement from uncertainty to revenue.

Nobody buys SEO because they emotionally crave backlinks.
Nobody buys paid media because they love campaign structures.
Nobody buys content because they admire blog formatting.

They buy these things because they are trying to move a buyer through a chain:
unaware → aware → interested → trusting → wanting → acting.

Every marketing function exists to move somebody forward in that chain. If it does not move somebody forward in that chain, it is probably just organizational theater disguised as marketing sophistication.

When I reduce marketing down to first principles, most of the work is actually just recombining a relatively stable set of raw materials:
customer pain,
customer language,
proof,
offers,
objections,
competitor claims,
search demand,
social demand,
conversion data,
sales feedback,
brand taste.

That is it.

Most campaigns, landing pages, ads, outbound sequences, webinars, sales decks, and SEO pages are simply different combinations of those inputs expressed through different channels.

The problem is not lack of intelligence. The problem is workflow fragmentation.

Traditional agencies route these inputs through an absurdly inefficient human maze. One person gathers data. Another writes a brief. Another reviews the brief. Another asks the client clarifying questions. Another creates variants. Another formats slides. Another builds reports. Another explains the reports in a meeting nobody wanted.

By the time the work actually ships, the market already moved.

This is why I increasingly think the future agency is fundamentally about loops, not services.

A loop is an end-to-end process that continuously converts inputs into outcomes while learning from feedback. The important part is not automation. The important part is the feedback system.

For example, take a paid creative loop.

Customer pain, proof points, offer positioning, and channel data become creative angles. Those angles become variants. The variants become ads. The ads generate performance data and sales feedback. The winners get scaled. The losers get killed. The learnings get stored. Then the next iteration improves.

That is a loop.

An agent is not the loop.

An agent is simply a worker operating inside the loop.

This distinction is where I think a lot of AI companies are going to fail. Right now everybody is obsessed with building agents. Research agents. SDR agents. Creative agents. Analytics agents. Reporting agents. Coding agents.

But many companies are building giant collections of agents without building the actual operating system that coordinates them.

So you end up with:
many bots,
many automations,
many dashboards,
many copilots,
and still no meaningful compounding advantage.

That is AI theater.

The future agency probably has three layers.

At the top sits the managed loop. That is the actual business outcome layer.

Underneath are specialist agents responsible for execution tasks.

And above everything sits human judgment.

I actually think humans become more valuable in this world, not less valuable. But their role changes significantly.

Humans should own:
taste,
strategy,
prioritization,
tradeoffs,
offer quality,
client trust,
narrative framing,
high-consequence decisions,
what to kill,
what to scale.

Humans should not spend their best cognitive hours resizing creative, formatting slides, manually checking broken links, rebuilding onboarding checklists, or writing generic first drafts for the 900th time.

That work should become infrastructure.

This is why I think codified workflows become incredibly important. At Single Grain and Single Brain, I increasingly think about this through SKILL.md files.

A skill file is basically codified expertise. It defines when a workflow should run, what inputs are required, what tools are involved, what good output looks like, what failure looks like, what requires human approval, and what gets saved back into memory.

Without codified skills, agents improvise.

With codified skills, systems begin repeating the best version of the workflow instead of reinventing it every time.

And that is where compounding starts happening.

Labor resets every month.
Infrastructure compounds.

I think the agencies that survive this transition will own seven major loops.

The raw materials loop continuously collects customer pain, objections, reviews, transcripts, competitor messaging, proof, search demand, and social signals so inputs never go stale.

The creative testing loop continuously generates, tests, kills, scales, and learns from creative velocity.

The demand capture loop watches search, AI recommendations, social discovery, and community conversations to determine where authority should be built.

The conversion loop identifies where momentum dies between impression and revenue.

The client narrative loop converts performance data into clear decisions instead of recurring meetings.

The sales enablement loop continuously feeds sales teams with better proof, better objection handling, and better follow-up assets.

And finally the learning loop captures every successful and failed pattern so the entire organization compounds instead of restarting from zero with every client.

That last loop is the moat.

If an agency serves 100 clients and still starts from scratch every time, that is not a learning organization. That is a staffing company with a better website.

The firms that win in the AI era will not simply “use AI.” Everybody will use AI.

The winners will build systems that learn faster than competitors.

That changes hiring too. The most valuable people in the next generation agency are probably not generic task executors. They are people who can direct systems, judge outputs, recognize weak signals, make decisions under ambiguity, improve workflows, and earn trust.

Some people become dramatically more valuable in this world.

Some roles become brutally exposed.

That is uncomfortable, but pretending otherwise does not change the economics.

The question I keep coming back to is simple:

If AI makes execution dramatically cheaper, what do clients still need agencies for?

I think the answer is:
ownership of the growth loop.

Not isolated tasks.
Not disconnected deliverables.
Not prettier reports.

They need someone to continuously turn messy market signals into reliable customer acquisition while the system keeps getting smarter over time.

That is the business model I think eventually captures the full $7.