How marketing actually works when one person runs the whole function.
written for the one person who has to do all of it
A working framework for running marketing when most of the execution is done by AI agents. Walks through the work in three layers (what humans decide, what agents execute, the plumbing that keeps the rest from blowing up), the eight workstreams where output actually happens, and the six shapes of agent you end up building around. Built from a year of working on it in the open and reading roughly four hundred primary sources so future readers don't have to. Released under CC-BY 4.0 because it's more useful that way.
Most marketing problems are one constraint in disguise.
Two minutes. Name the single thing throttling growth right now, see the revenue-justified order to fix it, and get routed to the play that does.
Nine plays, one per constraint.
Each play decides what to do and why, ties the move to revenue, then routes the execution into the agent fleet. Start with whichever one the diagnostic hands you.
ICP & Jobs-to-be-Done
Define who you're for by the job they hire you to do, not by firmographics alone.
Positioning
Win the comparison the buyer is actually making, starting from their alternatives, not your features.
Category Design
Decide whether to enter an existing category or design a new one. Usually you enter.
Pricing & Packaging
Charge by a value metric your buyer's finance team can defend, then fence good-better-best.
GTM Motion
Pick one primary motion and let it dictate pricing, hiring, and the shape of the funnel.
Growth & Loops
Find the one loop that compounds and the single constraint throttling it.
Brand
Build mental availability against the buying situations, with assets that are actually distinctive.
Measurement & Attribution
Triangulate MMM, incrementality, and self-reported attribution. Distrust last-click.
Content
Tie every content gap to a service and the demand it serves, or kill it.
Eight domains, run with agents.
The agentic "how" under every play. Each domain has its own page: what the work is, why it matters now, the tools teams actually use, what tends to go wrong, and the metrics that matter.
Sensing & Intelligence
Convert raw activity into structured signal feeds.
Strategy & Positioning
What to be in the market: the most agent-resistant domain.
Content & Creative Production
Highest velocity, highest brand-risk domain.
Distribution & Channel Operations
Where content meets audience, including the 83% in dark social.
AI Search & Answer Visibility
AEO/GEO/LLMO, the newest standalone discipline.
Demand & Conversational Pipeline
What turns attention into revenue.
Customer Intelligence & Synthetic Testing
Pre-launch validation, audience simulation, digital twins.
Measurement & Attribution
The mirror; without it, the rest is theater.
See where you actually are.
click any cell to open it
Pick any of the eight domains. The matrix shows four levels of capability (Crawl, Walk, Run, Fly) with named teams at each level, the tools they use, the next move to advance, and the pitfall that catches most people at that stage.
Open the matrixManual baseline, no AI assist
Single-task copilots in the loop
Multi-step agents, human review
Autonomous fleet, exception-only
Ten things the field is still figuring out.
Each one with where it stands today, what to watch for, and a practical move you can make this quarter without waiting for the answer.
- 01Agent-to-agent commerce + protocols
- 02Brand-to-LLM communication
- 03The Ghost Workforce labor problem
- 04Regulated-industry synthetic methodology
- 05AI-Agent Buyer Behavior
- 06The Accountability Question
- 07Saturation / Commoditization
- 08Sentiment Risk in AI Search
- 09Wikipedia Editorial Wars
- 10The Creative Quality Ceiling
Sourced, not made up.
- Every statistic has a name and a date next to it. Conductor 2026, Jasper 2026, McKinsey, Ahrefs, IAB, peer-reviewed papers. If a stat shows up without a source, it's been flagged.
- Case studies are named, with real numbers. Not "a SaaS company saw lift" but Vercel going from less than 1% to 10% of new signups via ChatGPT, Rootly from 3% to 30% citation rate, SAP's "Digital Chop Shop" doing $23M.
- Honest about what couldn't be verified. The "only 2% of AI SDR implementations survive past year one" claim is flagged as unverifiable. The widely-cited 94% accuracy stat from the Park synthetic-research paper is corrected to 85% (the actual number in the paper).
- Updated every quarter. The open questions section gets the most churn because the field is moving fastest there. The full v4 plan is published in the open in
research-plan.md.
Want this run for you?
I run this operating model as a fractional CMO: one operator, an agent fleet, and the judgment to delete eighty percent of what the agents make. If one of these constraints is yours, let's talk.
Prefer the full form? Reach out here.