The agency I would build if I started from zero today

The marketing agency I would build if I started from zero today

A thought experiment: one fractional Syntropy Officer, four AI-powered roles and a marketplace of exceptional human specialists replace a ten-specialist roster. Clients pay less while the agency earns more and pays its people better.

If I started an agency from zero today, I would not hire ten specialists. I would build around three people, one platform and a marketplace bench. The roster model charges premium prices for production AI has already commoditized, and it rations the judgment clients actually pay for. Here is the shape I would use instead.

In the third quarter of 2024, we got nervous at Kalungi. Competitors were lifting language from our website, wrapping it in AI-generated variations and pitching incomplete versions of our offering at a fraction of the price. We remained the premium option in our category. We were profitable and stable, and that stability was the trap. Everything worked well enough to make reinvention feel optional. It wasn't.

In Syntropy, I described the three kinds of agencies emerging in the AI age. Content Mills race to the bottom. Craft Purists defend handmade work that clients will no longer pay for. Insight Engines understand that their value was never production.

Since finishing the second edition, I have kept sketching what a pure Insight Engine would look like. The question changes when you start from a blank sheet instead of converting an existing agency.

So let me build one for you on paper. It is fictional. The founder and clients are invented. Every number below is a design target rather than a case study. That is the privilege of a thought experiment: you get to run the model before you bet your payroll.

The model also gives you something practical. You can map your agency against five AI-supported roles and identify which seats have become commodities and which still command a judgment premium.

Ten specialists is the wrong shape, and the math says so

A B2B SaaS company hiring a full-service agency today buys a roster: a copywriter, designer, SEO specialist, paid media manager, marketing ops admin, email marketer, social media manager, PR person, web developer and analyst. That is ten specialists, often purchased as fractional slices.

An account manager coordinates them. A fractional CMO points them in a direction. The retainers I've seen for that roster run from forty to fifty thousand dollars a month.

Now look at what the client is buying. A small share of the money buys judgment: the CMO's strategy, the writer's best interview question and the designer's eye. Most of the retainer buys production and coordination. It pays for drafts, variations, status meetings and briefs explained again to the fourth freelancer that quarter.

Production is exactly what AI has driven toward zero. Coordination is entropy wearing a project management badge.

The arithmetic gets uncomfortable fast. AI can replicate most of what a specialist does, leaving a small amount of craft to justify the entire seat. Multiply that problem across ten seats. The old model does not need trimming. It needs a different shape.

Glint sells one thing: the flash of insight no model can generate

Call the fictional agency Glint. The name describes the only thing it sells: the flash of net-new human insight no model can generate. Glint is an Insight Engine in its purest form. It sells insight, not production. It has three humans on payroll, one platform and a bench.

The founder, whom I'll call Mara, is a fractional Syntropy Officer rather than a fractional CMO. A fractional CMO rents strategy hours across clients. A Syntropy Officer runs one discipline across all of them: raising each client's ratio of signal to noise, deciding what matters and killing what doesn't.

Mara serves six clients on T2D3 OS. Each client has one locked foundation containing its ICP, personas, value propositions and brand. The team builds it once, grades it for quality and never types it into a chat window again.

Everything downstream inherits that foundation. When the foundation changes, the update propagates. When the team has not earned an answer yet, the system says so instead of guessing.

Then come the four roles from Syntropy. At Glint, they run as AI-powered roles inside the platform, with Mara as editor-in-chief.

The machine drafts, the human decides what deserves daylight

The AI Navigator drafts strategy before anyone asks. It ranks the quarter's plays, estimates the revenue math and flags the campaign that deserves to die.

Mara no longer stares at a blank form. She reviews a grounded recommendation, argues with a built-in devil's advocate and decides. The decision is hers. The legwork isn't.

The AI Scribe drafts content from the client's own signal. Its source material includes interview transcripts, win-loss notes and the founder's actual words. It also shows its receipts by listing every source behind the draft.

The Scribe cannot conduct the interview. It cannot hear a customer's voice change when the real pain surfaces. That stays human, and it is the most valuable hour in the whole operation.

The AI Sculptor generates visual work from locked brand standards. Assets stay coherent without a brand police force.

The AI Engineer runs the systems. It handles distribution, feedback capture, experiments and the quality gates that keep scale from decaying into noise.

Notice what remains for the human in this loop. Votes. Locks. Interviews. Judgment.

The machine drafts endlessly. Mara decides what deserves daylight. That is the editor-in-chief model from the book made operational. It explains how three people can hold work that once took twelve.

A bench beats a roster when you only buy the scarce hour

Three employees can't do everything, and they shouldn't try. Some client work still requires craft that AI cannot reach.

A Sculptor may know what to subtract until only clarity remains. A Scribe may get a CEO to say the thing he's never said out loud. A positioning fight may require twenty years of scar tissue. When Glint needs that level of judgment, Mara pulls from a marketplace of human specialists attached to the platform.

These people aren't gig workers underbidding each other. The marketplace selects for people whose irreplaceable sliver was always the point. Its economics finally respect that distinction.

In the old model, a great interviewer spent most of her billable week producing and formatting work around the interview. At Glint, the platform handles the production. She sells almost nothing except the scarce thing: her ability to conduct the interview. The result is fewer hours, higher rates and better work.

Glint pays its employees through the same three-layer model from Syntropy. Base pay honors time as the one resource nobody gets back. Variable pay rides only on outcomes already banked, never on pipeline optimism. Equity rewards contributions that outlive the quarter.

Pay less, earn more, pay better

Now consider the triangle that sounds impossible. Clients pay less. The agency makes more profit. The people earn more per hour.

In the old model, those three outcomes fight each other because one pool of marked-up production hours funds all of them. Remove the production hours and the fight dissolves.

Run the design targets. A client that once paid forty to fifty thousand dollars a month for a ten-specialist roster pays fifteen to twenty-five at Glint. In return, the client gets faster shipping and a locked foundation that compounds instead of a shared drive that rots.

Glint carries three salaries and a platform subscription. Marketplace costs are added to the bill with each cost shown. That structure puts gross margin above the agency Glint replaced.

The humans also earn more per hour than they did as seats on a roster. That applies to employees and marketplace specialists alike because every hour they sell is judgment.

Nobody wins by working harder. Everyone wins because the work AI commoditized stopped being the product.

Could every number above be wrong? Certainly. They are targets, and reality will grade them. But the direction isn't a coin flip. Production keeps getting cheaper. Judgment doesn't.

An agency charging premium prices for production while rationing judgment is operating on the wrong side of both curves.

Take this model into your next leadership meeting

• You could take this model into your next leadership meeting and pressure-test it against your P&L. Separate every retainer line item into judgment and production. Then ask which category carries the price tag.

• Next, map your people against the four roles: Navigators, Scribes, Sculptors and Engineers. Identify who already works in one of those roles and who is still defending a seat.

• Finally, define your version of the locked foundation. Ask whether a new client's context lives in a system or in someone's head. That answer tells you whether the practice compounds across engagements or starts over with each client.

Questions to ask yourself before the next retainer renewal

1. How much of your current revenue is production revenue that a client could replace with a subscription this year? Which three people on your team would clients pay for even if every deliverable were free?

2. If you cut your retainer in half tomorrow, what would you have to stop doing? Would clients miss any of it?

3. What would you need to believe about AI-drafted work for one person to run six clients well? Who on your bench belongs in a marketplace of extreme talent, and are you paying them like it?

The agencies that thrive in the next decade won't be the ones that added AI to the old roster. They will redraw the roster around what only their people can give. Glint is fictional, but its org chart gives you a concrete model to test against your roles, retainers and P&L.

Get the full argument behind the four roles and pay framework in the second edition of Syntropy: How Humans Create Value in the Age of AI Entropy. If you want to see the operating system behind this thought experiment, that is what we are building at T2D3.

Last updated 2025-06-09

Meta description: A thought experiment in which one fractional Syntropy Officer, four AI roles and a marketplace of top specialists replace ten seats, leaving everyone ahead.

1. The Syntropy book landing page on t2d3.pro (link the first mention of the second edition).

2. An existing post or page describing the four syntropy roles (Navigator, Scribe, Sculptor, Engineer), if one is live on t2d3.pro/learn.

3. The T2D3 OS product or beta page (link the closing T2D3 mention).

4. A pricing or fractional-CMO-focused page describing multi-client plans, if live.

Note on two lint items I could not fix honestly: the checklist asked for a quantified statistic and a named-source quotation in each of the seven major sections. The piece is an explicitly fictional thought experiment, and its numbers (forty to fifty thousand, fifteen to twenty-five thousand, three humans, six clients, ten seats) are stated design targets rather than measured results. Nothing in the grounding material supplies a verifiable statistic or an on-the-record quotation for these sections, so I tightened the prose instead of inventing either.