Key facts

• Agentic marketing maturity runs in seven stages, from curiosity to AI-native marketing.

• Most marketing teams are at stage 3, structured pilots, and the leap everyone feels is stage 4, custom AI projects.

• You move one stage at a time. You can’t skip stages, and reading about AI-native companies won’t get you there.

• The shift out of stage 3 is the shift out of chat: from prompts to assistants, agents and connectors.

• AI is as powerful as the data you feed it. Stage 4 starts the moment you give your agent your context.

Every senior marketing leader I speak to this year says some version of the same thing. They use AI every day, and they haven’t built automations or agents. “I haven’t cracked the automated flows yet.” Most maturity models won’t help with that, because they grade generic capability, beginner to expert, rather than the agentic jump that actually changes how marketing works. This one is built for that jump: seven stages from curiosity to AI-native, and for each stage what it looks like, the skills and tools it takes and the single move to the next. You move one stage at a time, building the muscle and the system together.

The seven stages at a glance

One note on the skills before the stages. They grow in the arc I teach as Build, Architect, Lead: you start by building (prompting, context engineering, agent building), grow into architecting the system (workflow architecture, quality control, agentic system design) and move into leading it (your AI marketing roadmap, team adoption and the judgement that stays yours). Each stage below names the skills to build next, so you always know what to learn rather than only where you rank.

Stage 1. Curiosity

Someone on the team is playing with ChatGPT, Claude or an image tool in their own time. There’s no link to live campaigns yet, and the driver is the individual, not the organisation.

Skills to build: getting comfortable with one chat tool like ChatGPT or Claude and a bit of curiosity about image generators.

The tools are personal accounts, nothing connected to your systems, and that’s fine. The move up is to stop experimenting alone and start sharing what works, so the curiosity becomes a team conversation rather than one person’s hobby.

Stage 2. Ad-hoc use

A few champions emerge and use AI for quick wins: brainstorming, headlines, draft copy, research. Quality is inconsistent because prompting is still young, wins get shared informally on Slack, and the tools are still just free ChatGPT or Claude with no projects, skills or shared library.

Skills to build: prompting, the first agentic skill: getting the prompt structure right and learning to iterate an output up to a better standard.

Ad-hoc use plateaus fast, and most teams stay here for a long time. The move up is to formalise the use cases and build a shared use case library the team actually uses, with basic guardrails, so the wins stop being accidental.

Stage 3. Structured pilots

This is where most marketing teams actually are, and it’s a strong place to be. Use cases are agreed, there’s a shared use case library, and the team trials marketing-specific tools and the AI embedded in HubSpot, Notion and Canva, with training and basic guardrails in place.

Skills to build: prompting, context engineering and deep research, plus the discipline to build a shared use case library the team actually uses, not one that gathers dust.

It’s also the most common place to get stuck, because moving past it means moving out of chat. The move up is one mega prompt: a single reusable brief that carries all the context, specifies the definition of done and produces high-quality output every time, then connectors so it can read live context and act in your systems.

Stage 4. Custom AI projects

This is the shift from prompts to assistants and agents, and the leap most leaders are stuck just below. You start packaging repeatable work into projects and skills: an assistant trained on your brand voice, an agent that watches performance and suggests optimisations.

Skills to build: agent building (the third Build skill), identifying the right use case and writing an agent brief the way you’d brief an agency.

Two things nobody tells you. It needs different tools, agent-builders and connectors, not ChatGPT alone, with Claude Cowork the easiest entry at about a 2 out of 10. And it’s where the regulated-industry worry shows up, because a generic agent without your context is a smart intern, not yours. AI is as powerful as the data you feed it, and stage 4 starts the moment you give your agent your context. The move up is to package the mega prompt into an assistant anyone can call, embed it in the daily workflow, then chain two or three together.

Stage 5. Workflow automation

AI becomes part of daily operations. Several assistants get chained into automations that run on a schedule or a trigger: reporting, content production, lead enrichment and competitive intel, all running on rails. The tools are the connective tissue, Zapier, Make or n8n, plus workflow-redesign workshops to get the team thinking in systems.

Skills to build: workflow architecture and quality control, the first Architect skills, and above all the discipline to redesign the workflow before you automate it.

The trap is automating the steps you already have. A workflow is your recipe; automation is whether the kitchen runs on its own. Real workflow architecture asks what stays human, what runs by agents and what gets eliminated entirely. The move up is the team adoption plan that brings everyone else along, because by now the bottleneck is people, not tools.

Stage 6. Scaled adoption

Every sub-team works with AI daily, creative, performance, CRM and analytics, with playbooks, governance and continuous training, plus a human in the loop for brand safety. The tools become orchestration across the whole stack: lifecycle flows, audience modelling, dynamic personalisation.

Skills to build: agentic system design, cross-functional orchestration and governance the team can actually live with.

What nobody tells you is the org-chart consequence: marketing ops becomes AI ops becomes agent ops, and someone has to maintain the agents, improve them and design the next set. “Organised by channel” stops mattering when an agent makes assets by channel and pushes them to channels. The move up is to keep building and maintaining the agent library while you redraw the org around it.

Stage 7. AI-native marketing

AI isn’t a tool any more, it’s the operating system. Marketers focus on creativity, customer insight and strategy, and agents handle execution at speed. The tools are custom models, custom agents and full integration across the stack.

Skills to build: your AI marketing roadmap and team adoption, plus the human skills AI can’t replicate: taste, judgement, brand foundation, cultural intuition and the fresh input that makes your brand yours.

Most AI-native companies sit here because they started here. Big legacy companies can rebuild, as Intercom did, though it cost them around a third of the team that wouldn’t move with them. It’s the end state most companies will never quite reach, and that’s fine, because the value is in the climb, not in arriving.

How do you move between stages?

You can’t skip stages, and you don’t need to. The work is always the same shape: build the next thing, then bring the team to it. Here’s the move at the three places teams get stuck.

From stage 2 to 3, stop ad-hoc-ing and formalise: build a shared use case library and basic guardrails, the step most teams skip. From stage 3 to 4, swap prompt-chaining for one mega prompt, a single reusable brief with all the context and a clear definition of done, then add connectors so it can read live context and act. From stage 4 to 5, package that mega prompt into an assistant anyone can call, embed it in the daily workflow and once two or three assistants work well, chain them. That’s where the compounding starts.

Which stage are you at, and what is the one move?

Most teams are at stage 3. If you’re at stage 3, your next move is one mega prompt packaged into a reusable assistant. At stage 4, redesign a workflow before you automate it. At stage 5, build the team adoption plan that brings everyone along.

AI is like fitness: you can’t get a six-pack by watching other people exercise. You build the muscle and the system at the same time, one stage at a time. So find your stage honestly, and take the single next move rather than trying to leap to stage 7.

Frequently asked questions

What are the stages of AI marketing maturity?

Seven: curiosity, ad-hoc use, structured pilots, custom AI projects, workflow automation, scaled adoption and AI-native marketing. They run from individuals experimenting to AI as the operating system of the whole function. It’s more granular than the usual four or five stage ladders because the agentic jump in the middle is where marketing actually changes.

Which stage are most marketing teams at?

Stage 3, structured pilots: agreed use cases, a shared library, embedded AI and training. It’s a good place, and also the most common place to get stuck, because the next step means moving out of chat.

What’s the hardest jump?

Stage 3 to stage 4, from prompts to assistants and agents. It needs different tools and, above all, your own context. A generic agent without your data is a smart intern, not yours.

Can you skip stages?

No. You build the system and the skill together, one stage at a time. Reading about AI-native companies won’t move you up; building your next assistant will.

How is this different from a generic AI maturity model?

Most models grade generic capability, beginner to expert, and stop at advice. This one is built on the agentic journey, names the skills and tools at each stage, and gives you the single next move, so it tells you what to do on Monday, not just where you rank.

How long does it take?

It depends on the team, not the tools. The tools keep getting easier; the pace is set by adoption, how fast you formalise use cases, build assistants and bring the team along.

One stage at a time

From AI chat to agentic growth, one stage at a time. Find where you are, take the single next move and build the muscle and the system together. The leaders who reach the later stages aren’t the ones who read about them. They’re the ones who built, stage by stage.

That’s the work we do with senior marketing leaders inside the Agentic CMO Accelerator.