What does an AI-native marketing organisation mean for the agencies that serve it? In short: the client keeps a very small team, runs its marketing on its own data and AI layers, and stops paying agencies for hours of execution. What it buys instead is the system, the data plumbing, the accuracy checks and the judgement that a model cannot supply.
That stopped being theory this month. Marketing Dive reported from an Advertising Week New York panel where Benoit Vatere, Liquid Death's chief media and digital commerce officer, described running media and retail-media optimisation on an AI operating system built with its agency, Power Digital Marketing. The brand has been through 17 versions of that system, called Omega, in about three weeks. The goal is to run "a pretty sizeable budget" with "only two people internally."
Two people. If you run an agency, that number deserves a slot at your next planning offsite.
What is an AI-native marketing organisation?
Power Digital's CEO, Jeff Mason, gave the panel a useful ladder. Most brands, he said, are AI-curious: they use frontier models such as ChatGPT and Claude with some of their own data. The next rung is AI-assisted: a bespoke intelligence layer that every employee can use, and that improves as it learns how the teams work. The top rung is AI-native: an orchestration layer on top, so the work itself runs inside custom applications.
Mason's view, as Marketing Dive reported it, is that frontier models alone no longer give anyone an edge. The moat is the intelligence and orchestration you build on top of your own data, and it deepens over time.
Liquid Death is a slightly unexpected poster child. Its own ad creative criticises AI data centres. Yet Vatere says AI keeps the billion-dollar drinks brand "lean and mean" and stops it spending heavily on non-working media.
What is Liquid Death actually buying from its agency?
Look closely and the agency has not disappeared. It has moved. In August, Power Digital announced its own AI ecosystem: Iris, an intelligence layer; Omega, the operating system; and Creative Affinity, a creative intelligence layer. Omega combines the agency's engineering with the client's data and processes.
The visible change is cadence. Retail-media campaigns that used to be reviewed weekly or twice a week are now optimised daily. Instead of using Slack, writing queries or asking his team for posts, Vatere says he can check what is happening himself and share it internally. His advice to other marketers: stop living in spreadsheets, and automate any task that takes more than five minutes.
So the client is not buying a media plan and a monthly deck. It is buying a machine that makes the monthly deck unnecessary, plus the people who built it and keep tuning it.
Here is how we read Mason's three stages from the agency side of the table.
| Client stage | What it looks like | What it stops buying from agencies | What it buys more of |
|---|---|---|---|
| AI-curious | Teams use ChatGPT or Claude with some of their own data | Little, yet. Expect pressure on rates for desk research and first drafts | Help choosing use cases, guardrails and training |
| AI-assisted | A bespoke intelligence layer every employee can query | Monthly reporting, routine analysis, status updates | Data centralisation, tuning, interpretation of what the layer surfaces |
| AI-native | An orchestration layer; work runs inside custom applications | Manual optimisation, versioning, most execution hours | System design and upkeep, output checking, ideas, cultural judgement |
Why does it feel as if the technology is deciding everything?
There is an old idea in sociology called technological determinism: the belief that technology drives how organisations and societies change, and people simply adapt. It is the mood of most agency AI conversations right now. The tools arrive, the client restructures, the agency reacts.
The Liquid Death story is a decent rebuttal. The technology did not choose two people. A marketing leader chose that. An agency chose to become the engineer of the system rather than a supplier to it. Same technology, different decisions, very different agency economics.
The principle we keep coming back to is this: in connecting data to humanity, clarity emerges from chaos. A daily-optimised retail-media machine produces more signal than any two people can read. Somebody still has to decide what it means, which signals matter, and when the numbers describe a real shift in people rather than a quirk in the feed. That somebody can be you.
Which agency services lose value, and which gain?
Start with the chores. The jobs most exposed are the ones nobody went into advertising to do: monthly reports, status meetings, budget tracking, manual resizing, competitor audits assembled by hand. In an AI-native client, those happen inside the system, continuously, and nobody will pay a retainer for them.
What gains value is less obvious but more defensible:
- Data centralisation. The panel's first piece of advice was to centralise data. Most clients' data still lives in a dozen places owned by different teams.
- Output checking. The panel also said AI agent outputs must be checked for accuracy. An agency that can audit a system's conclusions becomes the client's second opinion.
- System design. Deciding what gets automated, in what order and under which rules is strategy work, and it is billable.
- Creative and cultural judgement. A model optimises towards what has already worked. It does not know what should exist next.
How should an independent agency prepare?
- Audit your retainers by task. List every recurring deliverable and mark which ones a two-person client team with a decent AI layer could do themselves within a year.
- Apply the five-minute rule to yourself first. Automate internal tasks that take more than five minutes before a client points out that you have not.
- Offer to centralise the client's data. It is unglamorous, everything else depends on it, and whoever does it becomes hard to replace.
- Build a verification habit. Make "how do we know this is right?" a standing agenda item, and trace AI conclusions back to the source data before they reach a client.
- Reprice around outcomes and systems. If the client's cadence moves to daily, a monthly retainer measured in hours makes sense to neither side.
- Decide build, buy or partner. Power Digital built its own stack. Most independents cannot fund an engineering team, which is where buying the capability comes in.
Do independents need to build their own AI operating system?
No, but they need what one provides. Mason's argument is that frontier models are now table stakes and the advantage sits in the layers above them. For a holding network or a well-funded digital agency, building those layers is a capital project. For a 40-person independent, it is a distraction from the work clients actually hire you for.
That gap is why the Defy platform exists. It gives independent agencies the same analytical and AI capability the holding networks have built, without the development cost or the maintenance burden: live competitor and audience intelligence, questioned inside Claude like a co-worker, with agents taking the repetitive delivery.
Whatever you choose, choose before your client does. An AI-native client will not wait for its agency to catch up. It will hire the one that already has.
