Search Atlas Founder On The Five Levels Of Agentic Marketing And Why Most Teams Stall At Two

Martech has been running an expensive workshop on how to write a good prompt for two years now. Everyone showed up, and plenty of ad accounts kept hemorrhaging money. The average marketing team came out the other side with a slightly better way to ask ChatGPT for a subject line. While the prompts improved, the underlying workflow stalled.

A decade inside enterprise marketing tells Manick Bhan exactly why so many teams are stuck. Bhan is the founder and CEO/CTO of Search Atlas, an AI-powered marketing platform that unifies SEO, paid media, and authority-building into a single system. Before Search Atlas, he worked as a quantitative analyst on Wall Street, founded LinkGraph, a digital marketing agency recognized three times on the Inc. 5000, and built OTTO SEO, one of the first autonomous AI agents for search.

"Marketing is changing in a way that it's never changed before, and it's moving very quickly, and yet the products they have in the market are really behind," Bhan says. Bhan puts the blame on an industry that sells transformation and ships a Slack notification. Marketers, he thinks, can feel the shift in their bones.

The horse everyone refuses to get off

Most teams upgraded their software and left their thinking where it was. They reach for AI the way they always reached for tactics, one task at a time, and call the busywork progress.

"Most teams are still just trying to figure out how to drop this prompt into ChatGPT or Claude, copy an output, and paste it into their CMS. And that's them using AI," Bhan says. "They're still on the horse. They haven't gotten off the horse to jump into a car. They're in the predecessor model of how this thing works, and they haven't fully upgraded their mental model."

Only 49% of martech tools are in active use, and just 15% of organizations qualify as high performers. The stack keeps growing while the work underneath it stays the same. That split is what one agency founder describes as the marketer's job shifting from doing the work to deciding how the work gets done.

Prompting is the distraction, the system is the point

The fixation on prompt engineering, to Bhan, is what thinking too small looks like. The real problem is what happens after the prompt. Marketing work still gets handed from one system to the next, with people filling the gaps in between.

"People are still thinking it's like a prompt, that they're a couple prompts away from trying to automate something with a script or some throwaway code," Bhan says. "They don't yet know how awesome it is when you market with all the data integrated and all the systems able to talk and interconnect to one another."

47% of martech buyers name stack complexity and integration as the thing blocking value. A website that cannot reach the ad accounts, analytics that never report back to the CMS: the fragmentation turns marketers into full-time reconcilers of their own dashboards. Until those systems can talk to each other, someone has to carry the information between them. AI is only as useful as the information it can actually reach.

Bhan has an obvious stake in that argument. Search Atlas is built around the connected system he is describing. Whether companies are actually changing how they work or just adding another layer of software is the open question.

Multiplayer in the loop

The usual choice is framed as full automation or keeping a person in the loop. Bhan argues for something closer to how marketing teams already work. "There's full automation, there's a human in the loop, and then there's multiplayer in the loop," Bhan says. "Marketing always was multiplayer. That's just not how we were able to run those campaigns. That's the first thing that needs to change."

Marketing was always a team sport played by operators stranded in separate platforms, none of the systems aware of the others. Search Atlas runs through Slack, so a whole team works in one place while the platform reaches into the systems underneath them. The shift is less about removing people than changing what they do. Instead of feeding a system one instruction at a time, the team steers work that moves across systems and keeps going after the meeting ends.

Bhan sees that progression as a series of levels:

1. A marketer prompts a model, copies the output, and pastes it into the CMS manually.

2. Teams string prompts together with scripts or lightweight automation, but every task still starts from scratch.

3. Systems are connected and share data, so the AI can actually reach what it needs to do the work.

4. The whole team steers campaigns from one interface while the platform moves work across systems underneath them.

5. The AI starts improving how the work gets done without needing a new instruction every time.

At the bottom, someone is still copying an answer out of ChatGPT, checking it, pasting it into another system, and starting the next task. At the higher levels, AI takes action, carries context between systems, reports back to the team, and eventually improves how the work gets done without needing a new instruction every time. "Most people are at level one or level two," Bhan says. The important jump isn't adding another AI tool. It's getting the work to move without someone manually pushing it from one place to another.

Start at the foundation, not the finish line

None of this works if the systems can't reach the data they need. A fragmented stack caps what an agent can know and do. Nothing draws Bhan's contempt faster than a marketer pouring budget into ads aimed at a broken landing page. The basic work, he argues, is the part everyone wants to skip.

"Most marketers are just trying to sprint to the finish line without even starting with the foundation," Bhan says. He means the CMS, the conversion tracking, the site speed, the plumbing that decides whether any downstream spend converts. Break those and AI just gives the machine more ways to move bad information around.

Bhan is just as hard on the vendor market. Generalist platforms wire an agent to a website and call it a product. "You're just basically a stupid bridge that Claude can kind of do on its own," Bhan says. The value is in giving the agent enough marketing context and expertise to do something useful once it gets there.

The uncomfortable middle of a shift

Bhan gets why this moment rattles people. He calls the current moment a hurricane, a stretch where telling hype from substance is hard and where booking a strategy meeting feels a lot like doing something.

"If you're at an organization that has strategy meetings about learning how to use AI and spends a bunch of time in a room talking about it and not actually doing it, you're in the wrong room," Bhan says. His prescription is simple: stop talking about it and start building.

Bhan holds his own team to it. Every employee has to code, and he runs weekly sessions where they watch him work and pick up patterns in real time. His head of HR, once allergic to spreadsheets, now builds her own dashboards and commits to repositories. The examples are specific to his company, but the lesson is broader: change the tools without changing what people are expected to do and the old organization stays intact.

Companies spent the first phase of the AI transition learning how to talk to models. The next phase is about changing how the work itself moves: who does what, what systems can see, and which tasks still need someone pushing them forward.