When I tell people my agency has no employees, I almost always get the same question: then who does the work? The honest answer sounds strange at first. Most of it is done by AI agents I built myself, and they run every day. The rest is done by me. That last part is what my clients are actually paying for.
I deliberately avoid calling this a cost-cutting model. If all you want is to squeeze costs, you have someone cheaply copy text templates somewhere and call it marketing. This is something different: an operating model where machines handle the volume and one person carries the responsibility. That changes what a small agency can deliver. And it changes the conversation with the client, because it is no longer about hours, it is about what actually runs at the end of the day.
What a normal workday looks like
In the morning I review what happened overnight. Agents have checked bids in the ad accounts and logged anything unusual. Another one has collected new Google reviews and prepared draft replies. Yet another has worked through the weekly content quota: copy, plus image suggestions and variants for each location. They write reports too, sober and backed by numbers, the way I would write them myself, except I do not spend two hours at my desk doing it.
My role in all this: I build these systems and I monitor them. And I answer for everything that goes out. Nothing reaches a client or the public without my sign-off. That is not a formality, that is the core of the model.
Here is an example that makes it concrete. One client, a medical practice group with seven locations, had accumulated 42 unanswered Google reviews. That happens faster than you would think: seven profiles in day-to-day operations, and nobody feels truly responsible. We cleared them in one day. Every reply written individually, matched to the content of the review, no copied boilerplate. And privacy-compliant, which is trickier in this industry than many people realize: a reply must not even confirm that someone was a patient at all, let alone discuss a treatment. The agents delivered the drafts, I checked and approved every single one. A person working alone would have needed several days for 42 individual, legally clean replies.
What the agents cannot do
Now for the part many vendors like to leave out. Agents do not take responsibility. If a reply to a review misses the mark, no model apologizes to the client. I do. That is why nothing gets published fully automatically at my agency, even though that has been technically possible for a while.
Client conversations also stay with me. A managing director weighing whether to promote a new location does not want a chatbot. They want someone who knows the numbers and the story behind them. You cannot delegate those conversations, not to employees and not to software.
And then there are the edge cases. In healthcare marketing you constantly run into phrasings that are just barely legal or just barely not. Germany’s medical advertising law (HWG) prohibits promises of results, and the line between “describes a treatment” and “promises an outcome” often comes down to two words. An agent can know the rules, and mine flag everything uncertain instead of guessing. But when it is a judgment call, I make it, because I know the client and I am the one who answers for it in the end. Judgment is not compute.
What the client is actually buying
To be fair, the other side belongs in this story too: without the agents, my offering in this form would simply be impossible. The content pipeline for that same practice group delivers 5-7 posts per week, distributed across four channels, plus localized variants for the seven locations. Do the quick math and you end up with a weekly volume of copy that no single person could write and publish manually. A traditional agency would put a small team on this, and the price would look accordingly.
At my agency, the raw material is produced by machines, following specifications I defined cleanly once: tone of voice, brand colors, legal guardrails, local references for each location. The pipeline stops at defined checkpoints and waits for approval. I read everything and release what holds up. My weekly effort stays manageable, the quality stays consistent. And when a specification changes, I change it in one place and every future run inherits it automatically. That is the real difference compared to a team of people: a system forgets nothing and does not have bad Mondays.
You could object: so clients are buying software. That is not quite right either. They are buying a running system plus the person who built it and stands next to it every day. That is a different product than personnel. Staff get sick or quit, and knowledge walks out the door with every departure. A system keeps running, and the knowledge lives in the specifications, not in heads that can leave.
For the managing director of a practice or clinic, this comes down to two things. First, speed: tasks that used to run in weekly cycles get done in days, sometimes the same day. Second, one direct point of contact, no detours. There is no account manager at my agency who takes notes and passes requests along internally, and no junior who needs to be trained first. Whoever calls talks to the person who built the system and is responsible for it.
Whether this model works for every agency, I do not know. It requires a founder who can build things himself and wants to stay in the engine room instead of delegating. But for the work I do, visibility and inquiries for medical practices and clinics, it is the more honest setup. Clients are not paying for a payroll. They are paying for something that runs. And when something jams, they get the person responsible directly on the phone, not that person’s assistant.