Founders pitching an agency that runs AI agents keep showing up with a tech deck. Wrong instinct. The slide that eats the most airtime in the room is the one that draws a line down the middle of a page and labels two columns: what the agents do, and what the humans still do.
Investors want to see the line. Clients want to see the line. Agencies that can draw it clearly are winning work. The ones waving their hands about "AI-powered" everything are not.
A lot of the talk around agent-run marketing is wrong in specific, correctable ways. Here are the assumptions worth retiring before the next planning cycle.
The Agents Are Not Replacing the Agency
The loud version of this story says AI agents will swallow the agency whole. What's happening on the ground is narrower: agents are swallowing the tasks that used to justify the billable hour, and the agency is reorganizing around the work that's left. Keyword clustering, draft outlines, outreach list building, rank tracking, status reports. Those get handed to agents first because they reward volume and speed. Strategy, brand voice calls, the decision about which fight to pick this quarter, those stay with people.
A useful framing from SAS describes a new division of labor in which agents handle structured execution and repetitive analysis while humans keep strategy, judgment, and accountability. That's also what the pitch-deck slide is showing: not a roadmap to zero humans, but a map of where the human signature still has to land.
Flat Fees Are Already Replacing the Hour
The common assumption is that pricing will change later, once the software "matures." It's already changing. When a report that used to take a junior analyst half a day comes out of an agent in five minutes, the line item can't keep pretending to be an hour.
Agencies that notice are repricing before clients force the conversation. Those that don't are watching margin slip while they charge for work software now does for free.
The shift tends to land in three shapes, and the pitch deck usually names them plainly:
- Per-deliverable pricing. A price per published article, per optimized landing page, per outreach sequence. The unit is the thing produced, not the hours behind it. This is the slot where Digital.Marketing's AI-agent model recognized by Adweek sits, an agency built around agent-produced deliverables with human oversight on top.
- Outcome-linked retainers. A base fee plus a shared stake in a defined result: qualified pipeline, booked revenue, a ranking threshold. Harder to sell, harder to measure, and the model clients increasingly ask for by name.
- Platform or supervision fees. A flat monthly charge for keeping the agents running, tuned, and audited. It's the line that used to be buried inside the hourly rate, pulled out and named.
Oversight Is Not a Formality
The temptation is to treat human review as a courtesy check, a signature at the bottom of something the agent already finished. The failure mode is specific and unforgiving: agents still hallucinate, and they do it fluently. The Mind2Web 2 benchmark found that even the best-performing agentic search system evaluated reached a hallucination rate of 23%, with other systems above 50%. In a marketing context that's a wrong statistic in a thought-leadership post, a misattributed quote in an outreach email, or a factual claim about a client's product that no one bothered to check.
Agencies doing this well have a named human on every workflow, a change log that records what the agent produced and what the human altered, and a short list of decisions the agent isn't allowed to make alone: anything regulated, anything about pricing, anything that goes out under a client's brand without a second set of eyes.
Judgment Did Not Get Easier
A common line in the pitch is that agents free strategists to "do higher-value work." Fair, but misleading if it suggests the higher-value work is somehow simpler.
The work left on the human side is the work that was already hardest: deciding which market to enter, which message is actually true, when a client's instinct is leading them off a cliff, when to push back. Agents will draft ten subject lines in a second. Picking the one that fits the brand this quarter is still a judgment call, and the person making it carries the weight of being wrong.
The Slide Is a Contract, Not a Diagram
Treat the "what the agents do, what we still do" slide as an org chart and it will age badly within a quarter. Treat it as a contract with the client and the investor, renewed every planning cycle, and it does real work.
It tells the client what they're paying for and who's answerable when something breaks. It tells the investor which part of the business scales with software and which part scales with hiring. And it forces the agency to answer, out loud, the question the whole industry is avoiding: when the agents get better, which side of the line does the next task move to?

