5 Levels of AI Marketing: From Prompts to Agent Teams

Why Most Marketers Are Stuck at Level 1

Shann Holmberg, founder of Lunar Strategy, built a multi-agent content system as the actual production workflow for his agency — not a side project. In the process, he identified five distinct maturity levels for AI marketing, and argues almost everyone is stuck at the bottom two.

The framework maps a clear progression from generic prompting to autonomous agent teams, with the critical insight that the gap between levels is taste and judgment, not technology.

Level 1 — Custom Prompts: “the output at this level is generic because the input is generic. one line of instruction produces one dimension of output.” Every session starts from zero with no brand memory, no audience context. This is where 90% of marketers stay forever.

Level 2 — Manual Skills: Detailed 1,200+ line prompt templates with frameworks (Schwartz awareness levels, proof stacking patterns, CTA structures). Output quality jumps dramatically, but “the ceiling here is your time. you can produce great individual pieces, but you can’t scale because every piece still needs your hands on the keyboard.”

Level 3 — Skills + Brand Foundation: The inflection point. A brand foundation file captures voice, tone, audience pain points, positioning, words you always/never use. Every skill references it automatically. “one foundation feeds all your skills. email skill, social media skill, landing page skill, ad copy skill. they all pull from the same source of truth.” This is where AI content stops reading as AI — because the system knows who “you” is.

Level 4 — Agents with Skills: One command triggers a full workflow. Research agent pulls data, idea agent structures concepts, writer agent produces drafts, critic agent scores them. Holmberg references @akshay_pachaar: “subagents are for decomposition — they break a big task into smaller independent pieces.” One person does the work of a small team with zero coordination cost.

Level 5 — Autonomous Agent Teams: The endgame. Agents share context, build on each other’s outputs, and compound knowledge over time. An 8-person team running 38 agents across 8 departments — content, SEO, paid ads, email, social, analytics, CRM, reporting — each with its own agent cluster sharing a central knowledge base.

6 Takeaways from the AI Marketing Maturity Framework

• The moat is taste, not tools — Anyone can set up agents; the hard part is having judgment worth encoding into them. Rick Rubin doesn’t play instruments on most albums he produces, but every album sounds intentional because his taste shapes every decision.
• Memory is the multiplier — Run 1 produces a brand voice document. Run 20 has audience feedback, personalized patterns per segment, and a track record of what converted. The system gets better every time it runs.
• Marketing scores 9/10 on AI exposure — Karpathy’s job exposure analysis shows market researchers, writers, editors, and graphic designers all score a 9. The jobs growing are the ones requiring judgment at levels 3-5.
• AI slop is a real signal — “AI slop” mentions grew from 461K to 2.4M in a single year (5x). AI ads show 19% better CTR, but purchase likelihood drops 33% when consumers know AI created the content. Brand foundation is the antidote.
• The ROI curve is exponential — Levels 1→2 takes an afternoon. Levels 2→3 takes a weekend. Levels 3→4 takes 1-2 weeks. Level 5 is ongoing infrastructure that compounds daily.
• The window is closing — Operating at level 3+ puts you in a tiny minority today. The compounding advantage of accumulated taste, judgment, and months of memory means late adopters can’t catch up.

What This Means for AI-Powered Organizations

Holmberg’s framework validates a core thesis: the value of AI agents scales not with the technology itself, but with the organizational knowledge encoded into them. Level 3 is the critical transition — the moment AI content becomes indistinguishable from human-crafted work because brand judgment is baked into the system, not applied after the fact. The teams running at level 5 aren’t just faster; they’re building a compounding intelligence moat that grows with every execution cycle.