Liquid Death built its brand by mocking the AI economy. In August 2026, the billion-dollar canned-water company was running ads that take aim at the data centers the whole industry depends on, as Marketing Dive reported at the time. Then, on October 5, 2026, at Advertising Week New York, its chief media and digital commerce officer went on stage to explain how the same brand is rebuilding its entire marketing operation to be AI-native. The contradiction is the story: satire on the outside, one of the most deliberate AI operating-model transformations you will hear described by a brand this size on the inside.
Speaking alongside Power Digital Marketing CEO Jeff Mason, Benoit Vatere, Liquid Death's chief media and digital commerce officer, framed the move bluntly: "It's being able to stay lean and mean. As a fast growing company, that's what matters. We want to be able to manage things efficiently, to not have to spend massive amount of dollars on non-working media." AI is the tool that lets a challenger brand move faster than competitors carrying larger organizations and more red tape. Marketing Dive published its account of the panel on October 7, 2026, and the model Mason laid out there deserves a wider audience, because it gives nervous marketing teams a map: three named phases between dabbling and transformation.
The ladder: AI-curious, AI-assisted, AI-native
Mason's framework sorts every marketing organization into three phases.
AI-curious organizations have democratized the use of frontier models like Claude and ChatGPT and added some structured and unstructured data to the model. "That's where I think most brands are today," Mason said, "and frankly, that's not going to be enough to keep up with brands like Liquid Death." Everyone has the same models; access alone confers no advantage.
AI-assisted organizations create their own intelligence layer: a bespoke model accessible to all employees that improves by learning how teams actually work and operate. The intelligence layer turns scattered institutional knowledge into a queryable asset.
AI-native organizations build an orchestration layer on top of the intelligence layer, so work gets done inside their own custom applications rather than in generic chat windows.
| Phase | What the org runs | What it looks like day to day | Why it stops working |
|---|---|---|---|
| AI-curious | Frontier models for all, plus some data | Individuals prompting ChatGPT or Claude ad hoc | Everyone has the same tools; no durable edge |
| AI-assisted | A bespoke intelligence layer all staff can query | Answers come from your data, context and history | Still human-triggered; execution stays manual |
| AI-native | An orchestration layer on top, inside custom apps | Campaigns optimized daily; tiny teams run sizeable budgets | It doesn't: the layer compounds |
What AI-native looks like in practice
For Liquid Death, the anatomy comes from Power Digital's ecosystem, announced in August 2026: an intelligence layer called Iris, an operating system called Omega, and a creative intelligence layer called Creative Affinity. The numbers Vatere shared on stage are the telling part. Liquid Death has produced 17 iterations of Omega in three weeks as it fine-tunes the system. "Instead of going on Slack or writing my own queries or asking my team to post something, I can go in and have a sense of what's going on and be able to share that with my internal team so we can act fast," he said. "We optimize our campaign on a daily basis, especially on the retail media side. Things we used to look at on a twice-a-week or weekly basis, we look at on a daily basis now."
And the end state is organizational, not technical: "On a pretty sizeable budget, how can I operate that with only two people internally? That's the goal."
That is the AI-native pitch in one sentence: the cadence of decisions compounds, and the team size needed to hold the cadence collapses.
The Vatere rules
Vatere's operating rules for the team are quotable because they are extreme enough to be useful:
- No spreadsheets. "I don't want to see any spreadsheets anymore. Don't do all this time-consuming work. If there's something that can be automated or takes more than five minutes to do, do it once to train or build some kind of automation."
- Challenge the agents. Getting data together remains critical, as does challenging AI agents to ensure their outputs are accurate, and pushing team members to use and improve the AI systems every day.
- Frontier models are table stakes, not an edge. That is Mason's argument: off-the-shelf models no longer provide a competitive advantage when every firm uses the same tools. "Start centralizing your data, contextualizing it, building it in a way that you can interact with it... because over time that becomes a moat and competitive advantage," he said. "That becomes your own bespoke IP and your own bespoke intelligence, and that will grow over time."
There is a governance undertone here that the panel touched but did not dwell on: daily optimization loops only stay honest if someone owns accuracy checks and the audit trail, which is the gap between an AI-native org and an automation accident waiting for a post-mortem.
How a mid-size brand climbs without an enterprise budget
The Liquid Death story involves a large agency's proprietary stack, and that is the part you should not copy. A mid-size brand can climb the same ladder with the software it already has:
- Get out of AI-curious quickly. Most teams stall here because access feels like progress. Democratize the models, but pair them with written norms: what data goes in, what gets checked, what never ships without review.
- Build a small intelligence layer, not a bespoke model. Centralize analytics, CRM, ad platforms and revenue data in one place, then point scheduled analysis agents at it in defined weekly questions instead of one-off Slack requests. The "bespoke intelligence" is the data plus the questions, not the weights.
- Orchestrate inside your own apps. The phase-three pattern is a thin internal application that turns the intelligence layer into actions: campaigns checked nightly, budgets flagged, creative variants ranked. That is custom software territory, and it is cheaper than it sounds because the scope is narrow by design.
- Keep the layer yours. Omega is Power Digital's IP. If your intelligence layer lives in an agency's platform, the moat you are building is the agency's. Own the pipeline, the data and the orchestration, even if you hire help to assemble them.
Two guardrails from this month's other coverage apply directly. The HBR argument we covered on designing new processes instead of automating old ones is what separates phase three from phase one with a bigger invoice. And Zoho's budget cuts and salary freeze are the cautionary macro story: AI cost discipline eventually arrives at every org, so build the native loops while the pressure is still a tailwind, not a crisis. The adoption-results gap that CMI documented among B2B marketers earlier this year is the same warning at the industry level.
Conclusion
Liquid Death is an uncomfortable benchmark for most marketing teams because it removes the usual excuses: a company famous for ridiculing the AI economy in its ads is nonetheless building one of the most deliberate AI operating models described on a public stage. The ladder is named, the phases are observable, and the discipline behind them (centralized data, daily optimization, a five-minute automation rule, bespoke intelligence layers) is available at smaller scale than most brands assume. The only phase you cannot skip is the one most brands are stuck in: handing the same frontier models to everyone and calling it transformation. If you want help designing the intelligence layer or the orchestration apps for your own climb, talk to us.
Read next:
