What does it mean to be an AI-native agency, and how fast should you get there? Being AI-native means AI is built into how the work flows, so your frontline experts benefit without having to think about it. "How fast" is the wrong question. The useful one is: where is the finish line, how long will it take, and what will it cost to get there?
Brands are already setting their finish lines. Marketing Dive reports that Liquid Death is running media and retail-media optimisation on an AI operating system built with the agency Power Digital Marketing. The brand has been through 17 versions of Omega, Power Digital's operating system, in about three weeks. The stated goal is to run "a pretty sizeable budget" with "only two people internally".
Seventeen versions in three weeks is fast. But look at what sits in front of the speed: a destination written down in plain words. A sizeable budget, two people. Every iteration can be judged against that. Most agencies chasing "AI-native" have the speed and none of the destination.
What does "AI-native" actually mean for an agency?
Plenty of agency leaders define it as "everyone on the team uses ChatGPT". An agency founder speaking at an AI adoption session in Singapore this week offered a better test. He has spent 18 months trying to make his own roughly 20-person agency AI-native, and he compared the real thing to walking through Changi Airport without a passport. No queue, no forms. You just walk through.
For an agency, that means the frontline experts, the planner, the copywriter, the account lead, shouldn't have to log into ChatGPT at all. The AI should already be inside the flow of their work.
By that test, an agency where everyone has a chatbot tab open is not AI-native. It is AI-adjacent. People are still copying, pasting and re-explaining the client to an empty box every morning.
Why do agencies confuse speed with progress?
Because the fear is real. When every week brings a new model and every feed brings a solo founder claiming to have replaced a department, standing still feels like decline. The Singapore founder described his own response as pure hare: new model, new tool, everyone on a code account, Fridays given over to building. What was missing was the finish line. Nobody had defined what they were racing towards.
The tortoise wins the fable not because slowness is a virtue, but because every step is deliberate. His reframe for agencies: define the finish line, how long it will take to get there and what resources you need to bring along.
Agencies feel this pressure more than most. Clients ask what you are doing with AI. New-business teams want something impressive to say in the pitch. The temptation is to have an exciting answer rather than a useful one.
What are the five levels of AI adoption?
The most useful slide in the talk was a staircase. As the diagram shows, it has five steps, and each one asks more of your context, your people and your engineering. The level names are his; the agency examples are ours.
| Level | What it looks like | An agency example |
|---|---|---|
| 1. Chat-based | Every input typed or pasted manually | A strategist pastes a brief into a chatbot and asks for creative territories |
| 2. Agent-lite | Saved custom prompts and projects | A PR team keeps a project per client with tone-of-voice rules and boilerplate |
| 3. True agentic | Individual agents in your system, with context established | An agent drafts the weekly media report from connected platform data |
| 4. The agent team | Agents share one source of truth and work together | A listening agent flags a competitor launch; a reporting agent adds it to the client update |
| 5. The function runs itself | People set direction, review and make the judgement calls | Routine monitoring and reporting run end to end; the team reviews and decides |
The jump from level 2 to level 3 is where agencies tend to stall. Below it, AI lives in individual habits. Above it, AI lives in your systems, which means someone has to build, connect and maintain them.
Does every agency need to reach level five?
No, and the talk was emphatic about it. If level 2 solves your bottleneck, stop there. The objective is progress, not speed.
That plays out very differently by agency type. A small PR consultancy whose bottleneck is first drafts may be well served at level 2, with well-built projects holding each client's voice. A media agency drowning in cross-platform reporting probably needs level 3 or 4, because its bottleneck is data moving between systems. An influencer agency managing hundreds of creators may need agents that share one record of who agreed what, and when. A creative shop whose bottleneck is the brief may not need to climb at all until the brief is fixed.
Climbing past the level you need is not ambition. It is maintenance you have signed up for without a reason.
Why does the rush keep happening?
Social Change Theory helps explain it. Industries change when technology, habits and institutions shift together. When technology races ahead of the norms and structures around it, the gap fills with anxiety, imitation and shortcuts. People adopt because others are adopting, not because they have decided where they are going.
In agencies the casualty is integrity: of the work, of the process, of the promise made to the client. Rushed AI produces confident output nobody has checked and workflows nobody can maintain. The insight behind this piece puts it simply: in the pursuit of speed, don't let integrity be the casualty.
What does real AI adoption cost an agency?
Less than you fear for seats, more than you think for everything else. The Singapore founder put a seat at somewhere between $20 and $100 per person. That is the cheap part. Real adoption, he argued, takes time, resources and engineers, and engineers are expensive.
That is the asymmetry independents face. Holding-company networks have spent millions on analytics, AI and intelligence stacks. Liquid Death built its system with an agency partner that brought the operating system. An independent of 20 or 60 people can't easily hire an AI team, and shouldn't build one just to keep up.
This is why Defy exists: to give independent agencies the same analytical and AI capability as the holding networks, without the build cost or the maintenance burden. You can see how the pieces fit together on the Defy platform. Buying capability lets you climb a level when the bottleneck demands it, instead of hiring for a staircase you may never need.
How do I set a finish line for AI adoption?
Think tortoise. Six deliberate steps, written down before anyone opens a new account.
- Name the bottleneck. One process, one place where the work gets stuck. Not "AI strategy".
- Write the finish line in business terms. "Pitch research ready in two days, not five", not "use AI in new business".
- Estimate the time honestly. Include learning, testing and the dip in output while people adjust.
- List the resources. People's hours, engineering help, tool budget, and who maintains it afterwards.
- Choose the lowest level that clears the bottleneck. Try level 2 first if it might work. Climb only when it doesn't.
- Set a review date. Check progress against the finish line, then stop, adjust or climb, deliberately.
