An agency run on AI agents: where strategy has to go back to a human
Agents carry production well. The problem is elsewhere: strategy starts looking the same across every client, and clients notice before you do.
Sound familiar?
You run a digital agency for small businesses with six clients. No human executors — the work is done by five agents: strategist, designer, copywriter, media buyer, analyst.
- The process has been running smoothly for six months and production is fine.
- Every role has a strategic half and an execution half, and right now agents hold both.
- The strategist produces recommendations that look sensible — and identical.
- You'd noticed, but put it down to clients being in similar niches.
The moment it usually breaks
A client comes to the review and says: 'this strategy is word for word what my competitor is running — are you working for them too?' And there's nothing to say, because an agent built it from the same template.
What you actually need
Not 'fire the agents and hire people back'. You need to know where an agent is genuinely stronger and where its output is templated by its nature.
- See which recommendations repeat across clients before a client sees it.
- Split the strategic half from the execution half inside each role, not across the agency.
- Keep agents on the work where they give speed without costing quality.
- Know what putting a human back into a specific point costs and what it buys.
How it's put together
BossForce sets up a workspace for this: agents with clear ownership, a goal with a measurable metric, and playbooks — processes that fire on an event.
This example is built on a model situation, not a real agency's client roster. The structure and format are what the product creates during setup.
The goal
Separate the strategic half from the execution half: return the first to a human working from agent-gathered material, leave the second entirely with agents.
- Overlaps between clients
- 3 of 6 → 0
- Strategy
- agent → human, on agent material
- Production
- stays with the agents
Who handles what
Gathers what a strategy gets built on: what the client's competitors are doing, how demand behaves, what you've already tried. Draws no conclusions — it brings material.
View →Compares a proposed recommendation against what you've already sold other clients and flags the matches. This is the check that wasn't there when a client recognised their competitor's strategy.
View →Copy, artwork, campaign management, reporting — against a strategy a human signed off. This is where agents really are faster, and there's no reason to take it back.
View →Integrations
HubSpot and AI agents
A CRM is only worth what people bother to type into it, and nobody bothers. An agent writes down how a call or an email actually ended, notices the deal that hasn't moved in three weeks, and puts the follow-up you meant to send in front of you.
View →MailchimpMailchimp and AI agents
Email is the one channel a business owns rather than rents — the list belongs to you, not to somebody else's algorithm. An agent takes on the part that keeps the newsletter from going out at all: writing it, keeping the segments honest, setting up what follows a purchase, and then telling you plainly what happened.
View →Google SheetsGoogle Sheets and AI agents
In a small business the spreadsheet is the system of record nobody admits to: the client list, the price list, the shift rota, the numbers a report gets built from. An agent reads it as data rather than as a grid — pulling what the weekly report needs, keeping the register current, and showing you the rows that don't agree with each other.
View →Processes that run themselves
A new client, or a strategy review for an existing one
Gather the material: competitors, demand, what's already been tried. Run it through the overlap check against other clients. Hand it to you — the decision is yours, not the agent's.
End of the reporting week
Pull the campaign numbers, flag variance, prepare a draft report with sourced figures. The explanation of the variance is left to you.
What happens, step by step
- Work through four competitors' public channels
- Set out how they differ from one another
- Attach what's already been tried with this client
- Compare against the six live strategies
- Flag two positioning matches
- Your decision on what to change
- Read the material from the research agent
- Look at the flagged overlaps
- Your positioning decision
How this got set up
Where you're still required
This case is about the boundary rather than delegation. It doesn't run between roles — it runs through the middle of each one.
- Positioning and the strategic idea are human. The agent supplies material and a check, not the decision: on similar input it produces similar output, and that's a property, not a fault.
- The overlap check changes nothing by itself — it shows you the match before the client finds it.
- The explanation of variance in a report is written by you: the client asks why, and an agent's answer won't do there.
- Deciding that a client isn't a fit for your service is also human. An agent will not turn down money.
What changes
- Overlaps between strategies surface to you in advance rather than at a review.
- The strategic half goes back to a human and the execution half stays with agents — a deliberate split, not a retreat.
- Material for a strategy is assembled in hours rather than days, so you work with it instead of hunting for it.
- Production loses no speed: nobody took it away.
FAQ
Are you suggesting agents shouldn't do strategy?
Not that they shouldn't, but that their role changes. Gathering material and checking for repetition is something an agent does better and faster than a person. The decision stays with you, because similar data produces a similar answer — and in strategy that's exactly what the client isn't paying for.
What if the clients really are in similar niches?
Then the overlap is justified, and the check will show that. The point isn't that strategies must differ — it's that you know about the match before the client does and can defend it deliberately.
How many clients can be run this way?
Your time on strategy becomes the ceiling, not production capacity. The practical value of the split is that you know exactly where your ceiling is and what raising it would cost.
Does this apply to an in-house team rather than an agency?
Yes, the mechanics are the same: execution scales with agents, while positioning and priorities stay with people. What changes is who you present the result to.
Related walkthroughs
All walkthroughs →How a freelance designer handles 5 clients instead of 2
The bottleneck isn't how fast you work. It's the hours lost hunting for the current version and assembling decks. Here's what comes off your plate, and what stays yours.
Read the walkthrough →Bookkeeping & outsourced financeHow a solo bookkeeper goes from 8 clients to 20 without hiring
The ceiling isn't your expertise — it's the hours that disappear into categorising receipts. Here's what can come off your plate, and what has to stay under your review.
Read the walkthrough →Enterprise & productGetting through quarterly review without the 1 a.m. spreadsheet session
No team under you, no headcount coming, and you're held to a department's output. Here's which parts of that load are mechanical enough to hand off — and which have to stay yours, especially in the room.
Read the walkthrough →