A CMO at a 220-person B2B SaaS told me at a Pavilion dinner in March: “We mandated AI tool adoption in October. By Q1 we had a 92% utilization dashboard and a 14% drop in first-draft approval rate. I have never spent more time editing in my life.”
She had done what every consultancy advised. She set KPIs around agent use. She bonused her Directors on rollout speed. She added “AI-first” to the team’s quarterly objectives. The dashboard turned green within a quarter, and the work got worse.
Andy Grove wrote in Only the Paranoid Survive (1996) that the leader’s most important job during a strategic inflection point is not to lay down a roadmap but to set the example for the new way of working. He did not have agents in mind. The principle holds, and right now it is exactly what most CMOs are missing.
There are two paths CMOs are taking right now to drive agentic adoption. Both are failing. There is a third path that works, and it does not look like either of the first two. This is not an argument against AI adoption — it is an argument against the two specific mechanisms most leaders are using to try to get there.
Path 1: Pressure. The compliance-theater path.
The first path is to mandate adoption — make agent usage a KPI, tie it to performance, set quarterly utilization targets. This is what every change-management framework will tell you to do, and for prior enterprise software waves it worked. CRM in the early 2010s. Marketing automation in the mid-2010s. The mandate was the move because the tool was the bottleneck.
It is not the move for agents. Agents are not a tool; they are a workflow change. When you mandate workflow change through KPIs, you get what every adoption mandate produces: compliance, not adoption. ICs run the agent because the dashboard counts. They run the agent on the easy work, where the agent’s draft is okay enough to ship as-is. They keep the hard work — the brand-defining work, the work that takes judgment — for themselves, because they know what your dashboard doesn’t: the agent’s draft on the hard work is not okay.
The mandate produces volume on the wrong things. Volume on launch emails, volume on social posts, volume on the kind of content that compounds gently regardless of who wrote it. No volume — actually, often less — on the announcement that needs to land exactly right, the positioning paragraph that has to read as your team and not as Medium.
The CMO at the Pavilion dinner saw a green dashboard, a flat brand-lift number, and a Director who started looking for another job. The dashboard wasn’t lying. The dashboard was measuring the wrong thing.
Path 2: Become the prompt engineer.
The second path is the senior IC path. The CMO starts using agents in their own work — drafts their board update with an AI agent, runs their competitive briefs through a research agent, prototypes positioning copy with a writing agent. They get good at it. They start sharing screenshots in All Hands. They tell their team: “This is what I did. You should do this.”
The team watches the CMO get good at agents. They do not get good at agents.
The reason is structural — and it is the same structural reason that positioning experts have been pointing at separately for two years: the leader’s positioning doesn’t travel to the team without a substrate the team’s tools can also read. What the CMO did — and got results from — is senior IC work with agents. The CMO knows the brand voice cold; they have eight years of context in their head; they edit on instinct. When they prompt the agent, the agent produces something close to brand because the CMO knows where to push it. When the team prompts the same agent without the same head-state, the agent produces something generic, and the team rewrites it back to the CMO’s standard, eating the rewrite tax that I named in the previous post in this series.
The CMO who becomes a senior IC with agents has built a self-portrait of an agent-using leader. They have not built an org that uses agents well. The two are not the same problem, and the second is not solved by being good at the first.
Path 3: Lead with agentic.
The third path is the one nobody is naming yet. It is the path where the CMO does not mandate agent use, and does not perform agent use, but instead makes their own leadership work — the voice and current direction the team’s drafts need — agent-native, and lets the team’s adoption follow from the substrate that emerges.
What is agent-native leadership?
Agent-native leadership is the practice of making the leader’s voice and current direction — strategic decisions, positioning calls, brand standards, banned-phrase lists, ICP segmentation — flow natively through the same substrate the team’s agents draw from. The leader stops carrying context in their head and starts carrying it in a place the team’s tools can read. Three properties:
- The leader’s decisions become reachable, not just announced. A positioning call doesn’t live in an offsite slide; it lives in the substrate the team’s agents read from. The agent’s next draft reflects it within a day, not a quarter.
- The substrate is the modeling. ICs adopt the agent because the substrate makes the agent’s draft good enough to ship most of the time. They don’t need a mandate; the work itself becomes faster.
- Adoption follows from the work, not the directive. The CMO doesn’t have to tell the team to use agents. The agent’s first draft is now usable — the feedback you’d give, already in the draft — so the team uses it.
One more property, load-bearing for whether this lands inside the org: the substrate is not a leader-to-IC broadcast. Function heads — the Director of PMM, the Director of Demand Gen, the Group PM running the PRD layer — curate at function scope. The CMO sets brand voice and current direction; the Director of PMM canonizes positioning for PMM; the Director of Demand Gen canonizes campaign voice. Same substrate, scoped permissions. The Director is not bypassed by the substrate — the Director is the function-level owner of it, and the function’s voice gets carried into every draft the function ships.
What this looks like on Monday
A CMO leading with agentic does specific, concrete things their team can observe:
- Their strategic decisions go INTO the substrate the same week they make them. When the CMO settles on a positioning change after a competitive bake-off, the substrate updates that day. The team’s agents reflect the new positioning in next-week’s drafts.
- Their brand voice canon is maintained in the substrate, not in their head. The four phrases their brand voice forbids, the three things the team is allowed to say about V2 that they aren’t allowed to say about V1, the segment-specific language for ICP-A vs. ICP-B — all of it lives in the substrate. When the CMO refines a voice rule, the refinement propagates.
- They run their own 1:1 prep through an agent that reads from the substrate. Not as a stunt. As a demonstration that the substrate makes their own leadership work faster — and the team can see that the same substrate makes their work faster.
- They measure adoption by drift-incidents declining, not by agent-utilization rising. A drift-incident is an agent output that diverged from team direction badly enough that an IC had to rewrite or retract. Drift-incidents going down means the substrate is working. Agent-utilization going up means the dashboard is working. These are two different measurement frames, and most CMOs are currently watching the wrong one.
Why this works where the others don’t
Pressure works on metrics you can mandate. It does not work on quality, because quality is what the IC keeps for themselves when the work has to be right.
The senior-IC path works on the leader’s own work. It does not transfer to the team’s work, because the team does not have the leader’s head-state — and a screenshot in an All Hands does not transfer head-state.
Leading with agentic works because it does not require the leader to mandate anything or to perform anything. It requires the leader to put the context — the substrate the agents need — in a place the team’s agents can also reach. The team adopts the agent because the agent’s draft is now usable. Adoption is a side-effect of the substrate, not a goal.
The measurement difference is load-bearing. The KPI dashboard measures agent-utilization-rate — how often the agent gets invoked. The substrate dashboard measures drift-incidents — how often the agent’s output diverged enough from team direction that an IC had to rewrite. The first metric rises under a mandate and means nothing. The second metric falls when the substrate is doing its job and means everything.
How to start
Two specific moves in the first 30 days. The ongoing cost is small but real: five to ten minutes a day, every day, of curation — yours, or a chief of staff’s on your behalf. Hide that cost and the substrate decays; name it and the substrate compounds.
- Audit where your decisions currently live. Offsite slides, Slack threads, the back of your head, three Notion docs — wherever they are, they are not yet a substrate. Pick the four-to-six load-bearing decisions the team would benefit most from your agents being able to read. Put those in one place.
- Run one of your own leadership tasks through that place, with an agent. Your next board update, your next quarterly positioning brief, your next All Hands. Use the substrate. Notice where it falls short for your own work — that’s where it would fall short for your team’s work too. Fix it, by adding what was missing.
This is the substrate move. Pomegranate is one way to make it practical — I’m building it as the leadership substrate the team’s agents draw from, so the leader’s voice and current direction are reachable by every agent that touches the team’s work. Without that substrate, every catch-up move costs another rewrite cycle. With it, the team adopts agents because the agent’s draft is finally usable.
The CMO from the Pavilion dinner is now six months into the third path. Her utilization dashboard isn’t green anymore — she dropped it. The Director who was looking for another job didn’t leave.
That is what leading with agentic looks like.