I spent last week at SaaStr talking to 26 SaaS operators — VPs of Product, CMOs, Heads of Revenue — and the same question came up in almost every conversation: what does my job actually become when my team is mostly orchestrating AI agents?

One VP of Product put it plainly. She had reviewed the PRDs her team produced with AI agents this quarter. They were well-structured, grammatically clean, and strategically incoherent — three of them laddered to a segment the company had deprioritized in March. Good manager. Productive agents. Drifting work.

She is not alone. That pattern — more output, less coherence — showed up in nearly every conversation I had that week. The teams are shipping more than ever. The work is not staying on strategy.

The honest answer is that nobody knows the new shape of management yet. But after 26 of these conversations in a single week, I see four archetypes emerging. Most leaders default to one without choosing it. The right answer is probably a mix — but the weighting matters enormously, and most leaders are weighting wrong.

1. The Master of Judgment

The first archetype is the leader as final editor. The team produces more than ever; the manager’s job is to catch what’s wrong before it ships.

This is the default. It’s where most leaders end up without thinking about it, because the volume of AI-assisted output creates an immediate review pressure that didn’t exist before. When your team produces three times the volume, someone has to verify three times the output. That someone is usually you.

The Master of Judgment is necessary. You cannot skip it. But as a primary operating mode, it has a ceiling: it scales with the manager’s time, not with the team’s capacity. Every new agent the team adopts creates another review stream. The manager becomes the bottleneck they were trying to eliminate.

A VP of Marketing who spends 60% of her week reviewing AI-assisted drafts is not leading. She is copy-editing at the speed of one human.

2. The AI Coach

Editorial watercolor illustration of a man in his early 40s standing at a whiteboard, gesturing while explaining something to a colleague seated nearby with a laptop. The whiteboard shows diagrams and notes. Warm ink linework on cream background.
The AI Coach: teaching the team to prompt better.

The second archetype is the leader as prompt engineer for the team. They run workshops on how to brief an agent. They share their personal ChatGPT configurations. They build prompt libraries and distribute them in Notion.

This is valuable — for about 90 days. The coaching produces a measurable uplift in agent utilization. The team starts using tools they were ignoring. The quality of prompts improves.

Then it plateaus. Because the fundamental constraint on AI output quality is not the prompt. It is the context behind the prompt. You can teach someone the perfect prompting pattern and they will still produce a generic draft if the agent does not know the team’s voice, the current positioning, or what the CEO decided on Tuesday.

The AI Coach teaches mechanics. Mechanics depreciate as models improve. What does not depreciate is the judgment that shapes what the mechanics produce.

3. The Human in the Room

Editorial watercolor illustration of a woman in her mid-30s sitting across from a colleague in a casual one-on-one setting. She is leaning forward, listening intently, with warmth in her expression. Plants and natural light in the background.
The Human in the Room: holding the team together through the transition.

The third archetype is the leader who leans into what AI cannot do — empathy, culture, team health, the conversations that happen between the tasks. They read the emotional temperature of the room. They protect people from burnout as the pace accelerates. They hold the human fabric of the team together while the machines handle the output.

This matters. Anyone who tells you it does not has not managed a team through a transition this fast. The psychological safety of a team adjusting to agent-augmented workflows is a real variable that determines whether adoption happens or whether people quietly route around the tools.

But here is the tension: the Human in the Room, as a primary mode, cedes the strategic output to the agents — and to whoever is shaping what those agents know. If the manager is focused on team health but nobody is shaping the context the agents work from, the agents produce plausible, confident, strategically wrong work. The team stays happy. The output drifts.

4. The Owner of Context

Editorial watercolor illustration of a man in a blue shirt standing at a large wall board, connecting documents, strategy cards, and decision notes with red thread. The board shows labeled cards for Goals, Decisions, and Customer segments. Warm office light.
The Owner of Context: maintaining the layer that makes every agent better.

The fourth archetype is the one I think matters most — and it is the one I see the fewest leaders actively choosing.

The Owner of Context is the leader whose primary job is making sure every agent the team uses has the right background: the current strategy, the current positioning, the decisions that were made this quarter, the phrases the brand has retired, the segment the company is now targeting, the metric that matters this month.

This is not prompt engineering. This is not editing. This is a different activity: maintaining the context layer that sits between the team’s decisions and the team’s tools. When that layer is current, agents produce work that is 80% done — the team finishes it rather than rewriting it. When that layer is stale or missing, agents produce work that looks good and reads wrong, and the Master of Judgment works overtime catching the errors the context would have prevented.

The Owner of Context is the only archetype that compounds. Every decision you put into the context layer makes every future draft better — across every agent, across every team member, across every tool. The Master of Judgment scales linearly with effort. The AI Coach’s value depreciates as models improve. The Human in the Room is essential but strategically neutral. The Owner of Context builds an asset that makes the entire team’s AI output better over time.

All four matter. The weighting is what kills you.

Most leaders I meet are running roughly 50% Master of Judgment, 20% Human in the Room, 20% AI Coach, and 10% Owner of Context. That weighting produces an exhausted reviewer, a team that is trained on yesterday’s prompt patterns, a culture that is intact, and a context layer that nobody maintains.

The weighting I would argue for: 40% Owner of Context, 25% Human in the Room, 25% Master of Judgment, 10% AI Coach. That weighting produces a context layer that is current, a team that trusts the tools because the tools actually work, less review because the drafts arrive closer to done, and enough coaching to keep the team sharp on mechanics as models change.

The difference is not abstract. The VP of Product from SaaStr — the one with three PRDs laddering to the wrong segment — was not reviewing badly. She was reviewing without a context layer that would have prevented the drift in the first place. The agent did not know the segment had changed. No amount of review fixes an input problem.

What this looks like in practice

Five to ten minutes a day. That is the curation cost of maintaining the context layer — your team’s voice and current direction, kept current, reachable by every AI tool your team uses. A chief of staff can do it on your behalf. An ops lead can own it. The point is that someone maintains the layer the agents read from, and that the layer reflects this week’s decisions, not last quarter’s.

That is what Pomegranate does: it captures the decisions your team makes, briefs every agent before it drafts, and gives the team a surface to review what agents produce against what was decided. The feedback you’d give, already in the draft.

Of the 26 operators I talked to at SaaStr, the ones whose teams were producing coherent AI-assisted work had one thing in common: someone was maintaining the context layer. Not always the leader — sometimes a chief of staff, sometimes an ops lead. But someone was keeping the substrate current, and the agents were reading from it.

The job of a manager is not changing because of AI. The job of a manager is changing because the context that used to live in the manager’s head now needs to live where the agents can read it. The Owner of Context is the role that makes the other three work.