For an agency, the AI question is not whether to adopt the tools — it is what you sell once every competitor has them. When execution that used to take a week takes an afternoon for everyone, the billable hour that funded the old model stops making sense, and the firm has to decide what it is actually worth paying for. This guide is the Agency-model lane: the commercial decisions, and a pointer to the deeper piece on each.
The pillar holds the shared definition of AI-native media operations for anyone new to the term. This lane assumes it and goes straight to the questions an owner or leader asks: what to sell, how to staff it, how to price it, and where the advantage lives once capability is commoditised.
The thread through all of it is that AI changes what an agency can charge for, not how much effort it should look like it is making. Firms that treat this as a technology upgrade keep their old service list and watch the margin leave. Firms that treat it as a portfolio decision — what to keep, change, and stop selling — get to redesign the business while they still have the leverage to do it.
What to sell, and what to stop selling
The founding decision is a portfolio one. The agency-design decision framework sorts an agency's services into three buckets: keep, change, and stop selling. Anything a client's own AI can now do in-house belongs in the third bucket; anything where judgment, accountability, or data ownership matters belongs in the first.
| Bucket | Test | Typical answer |
|---|---|---|
| Keep | Human judgment, accountability, or proprietary data decides the outcome | Strategy, final creative judgment, data moat |
| Change | Valuable but reshaped — more volume, faster, priced differently | Production, reporting, localization |
| Stop selling | A client's own tools now do it at acceptable quality | Assembly work, format resizing, basic reporting |
What an agency should stop selling takes the third bucket seriously, because that is where denial is most expensive. Defending a service the client can now do themselves is a slow way to lose the relationship. The honest move is to name what is commoditised and move the conversation up to what is not.
The reason this is a portfolio decision and not a technology one is that the services are connected. Stopping assembly work frees the people and attention to staff the judgment work, and it changes how the firm looks to a buyer. A leadership team that cannot name its third bucket is usually still organising around producing the work — which is precisely the work AI has made cheap.
How an AI-native agency compares
AI-native versus traditional agency is a structural comparison, not a contest of enthusiasm. The traditional firm bills for execution hours and organises around producing it; the AI-native firm organises around reviewing and deciding, with AI producing the first pass. Different cost base, different shape, different thing the client is buying.
Is an AI-native agency cheaper? is the question every prospect asks, and the answer is less obvious than yes. The production cost is lower, but the valuable work — judgment, accountability, the data relationship — does not get cheaper. Some work gets cheaper; some work gets repositioned. Quoting a blanket discount is a way to signal that your work was assembly all along.
The practical response is to be specific about which line falls and which holds. A prospect comparing a lower total figure against a traditional firm's retainer is really comparing two different scopes, and the agency that explains the difference — rather than matching the number — is the one that keeps its margin.
What stays in-house versus agency is the client's mirror of the same decision. As the cost of execution falls, more of it moves in-house, and the split that survives is the one built around what the client cannot easily build: senior judgment, cross-client pattern recognition, and the memory your firm holds. That is a more defensible position than competing on production capacity.
Building and staffing the firm
How many people an AI-native agency needs is the headcount question, and the honest answer depends on the mix rather than a benchmark. Fewer producers, more senior reviewers, and at least one person who owns the data and workflow layer — because that layer is now load-bearing. Cutting producers without adding the operating role is how agencies end up shipping faster and worse.
The mix is the strategy, so it differs by what the firm sells. A firm that keeps strategy and judgment needs senior people and few of them; a firm that competes on volume still needs a review function it can defend. What it should not do is keep the old pyramid and layer AI on top, because the middle of that pyramid was production, and production is exactly what moved.
How to start an AI-native agency is the from-scratch version, and starting native is genuinely easier than converting a traditional firm, because there is no legacy cost structure to defend. The order matters: pick the position, build the client-memory layer, then add delivery capacity around it. Most new firms invert that and buy tools first.
Pricing and the moat
How to price AI agency services is the question that most determines whether the efficiency gain reaches the firm or the client's procurement team. The structural options — retainers, outcome-based fees, productised subscriptions — are not new, but AI changes which ones fit. Pricing by the hour in a world where hours collapse is a decision to shrink.
Building a client-memory data moat is what makes that pricing defensible. When every agency has the same models, the advantage is the accumulated context: what ran, what it cost, what it returned, and what the brand learned. Models are rented; that memory is owned. It is the one asset a competitor cannot copy by subscribing to the same software, and the reason AI memory matters more than model choice.
It comes with a condition worth stating plainly: the memory is the client's data, held in trust, not the agency's to reuse. A moat built without clear consent and boundaries is a liability dressed as an asset, and it is the fastest way to turn a data advantage into a trust problem.
Winning the work
AI for agency new business applies the same operating discipline to the pitch. Research, credential tailoring, and first-pass thinking can be produced faster, but the win still comes from demonstrating judgment the prospect does not have in-house. AI makes a firm look prolific; only the judgment makes it look worth hiring.
That is a useful filter for new-business effort. If a pitch asset could have been produced by any competitor with the same models, it is table stakes, not differentiation. The parts worth investing in are the ones that show what the firm noticed and decided — the same judgment the pricing section is asking clients to pay for.
Where this lane lands
The Agency-model lane is the commercial one, and it usually starts with a single decision: what to stop selling. Its natural home is the agency landing, which frames that decision — what to automate, what to keep human, what to charge for, and how to restructure around the answer.
If this is your bottleneck, the AI-Native Media Operations course walks the operating model and the agency blueprint in one place. The pillar holds the definition; the Strategy guide sets direction above delivery, and the Execution guide covers the workflow decisions the model rests on.
FAQ
Does the AI-native model mean charging less?
Not uniformly. The work that was assembly can and should be priced differently. The work that is judgment, accountability, and data ownership should not be discounted, because there is no reason it got cheaper. Blanket discounts usually mean an agency has not decided which half of its work it is actually selling.
Can a traditional agency convert, or is it better to start fresh?
Both are possible. Starting fresh is structurally easier because there is no billable-hour model to defend. Conversion works when leadership accepts that the change is a portfolio decision — what to stop selling — before it is a technology project, and this is the argument behind the decision framework.
Where should a small agency start?
With the client-memory layer and one service you can deliver AI-native end to end. Do not buy a stack first. The 90-day rollout applies here too: one service, measured, then expanded.
The short version
For agencies, AI turns a capability question into a portfolio question. Decide what to keep, change, and stop selling; staffing and pricing follow from that answer; and the durable advantage sits in the client memory you own rather than the models everyone rents. The tools will keep changing. The position will not.
I am still working this out in public. I have spent most of my career on the agency and client side of media, and the last few years rebuilding that work with AI, and I do not think anyone has the final answer yet. If your firm is somewhere on this path, I would like to hear which decision is the one holding you up.
Cheers, Chandler