What a growth team looks like in the AI era

Guide · Founder · 5 min read · last verified 2026-07-27

A growth team in the AI era organises around judgment rather than production. For a long time the shape of a marketing team followed the shape of its channels — an owner for search, an owner for email, an owner for social, each defined by the surface they operated. As agents absorb drafting, monitoring and research collection across every channel at once, the scarce activities shift: someone has to decide what the evidence means, someone has to decide what is good, and someone has to decide what ships. Those three decisions — reading, taste, and approval — are becoming the load-bearing roles. What follows is an org-design essay: observed direction and reasoning, not survey findings, and deliberately not an argument about headcount in either direction.

Production stopped being the constraint

The channel-owner org made sense when each channel demanded hand skills that took years to build. The constraint that justified it — producing competent work is slow and expensive — is the constraint agents dissolve first. First drafts, keyword-shaped variants, monitoring sweeps, research gathering, reporting assembly: these no longer queue behind a specialist's calendar. What teams discover when the production queue empties is uncomfortable and clarifying at once: the real bottleneck was never writing. It was deciding what was worth writing, whether the result was good enough to carry the brand's name, and whether anyone was allowed to publish it. The queue did not disappear. It moved to the decisions.

From channel owners to decision roles

A second dissolving force works alongside the first: evidence used to arrive channel by channel, in each specialist's separate dashboard, which made the channel owner the only person who could interpret their slice. Market growth intelligence platforms collapse that — a single research pass now covers buyer questions, competitor presence and AI-assistant answers in one evidence file; Magrios is built around producing exactly that artifact. When the evidence arrives unified, the org question stops being who owns which channel and becomes who reads the file, who decides what it demands, and who approves what goes out. Three roles fall out of that question. (The artifact and the category behind it are covered in this primer on the platform category.)

The evidence reader

The reader owns the map of loss: where buyers are asking questions the company never answers, where competitors surface and the company does not, what AI answers claim and get wrong. The craft is scepticism applied in two directions — against the sources, which must be graded rather than trusted, and against the team's own comfort, because unread evidence bends toward whatever the team already wanted to do. The reader's output is not a summary; it is a prioritised set of named gaps that the rest of the team can act on without re-deriving the research. The nearest current ancestors are the analyst and the strategist, but the centre of gravity moves: less collecting, which agents do, and more concluding, which they should not be left to do alone. The reader is also the role that keeps the team's output aimed — what separates a report from a plan is mostly whether this role exists and is heard.

The editor

The editor owns quality in a world where volume is nearly free. When anyone can generate competent material on demand, competence stops differentiating; the market fills with plausible, interchangeable content, and distinctiveness becomes the only scarce property left. The editor's job is therefore mostly refusal: killing the plausible-but-generic, protecting the voice, deciding which of many possible pieces deserves to exist at all. This is a promotion for taste — from a nice-to-have trait scattered across a team to a named, full-time accountability. The nearest ancestor is the content lead, but the ratio inverts: less producing, more judging, and a veto that the rest of the system is built to respect.

The governor

The governor owns the boundary between what agents may do autonomously and what requires a human signature. Claims discipline, brand safety, legal exposure, tone in sensitive territory — someone must define, in advance, which categories of output ship unreviewed, which need the editor, and which need sign-off above the team. Without this role, human-in-the-loop is a slogan that degrades into rubber-stamping under deadline pressure; with it, the loop is an actual design with named checkpoints. The governor also owns the failure drill: when an agent publishes something wrong, the question is not whose agent it was but which checkpoint was missing. This role has no clean ancestor in the marketing org — it borrows from editorial standards desks and from engineering's release management, and its existence is the clearest single marker that a team has moved from experimenting with agents to operating them.

What stays human, and why that is not a slogan

Four things resist delegation for structural reasons rather than sentimental ones. Framing the questions: agents answer well but choose questions poorly, and the choice of question is where strategy actually lives. Accountability: an approval is only meaningful when a person can be answerable for it, which is why the governor cannot be an agent. Relationships: partnerships, analyst conversations, customer trust — these accrue to people, not systems. And taste: not the ability to recognise good work, which models increasingly have, but the willingness to reject acceptable work in favour of a specific point of view, which is a commitment rather than a capability. Teams should expect the mix of their week to shift toward these four — that is the observable direction; its speed will vary by company and is not worth pretending to know.

A working shape for a small team

On a small team these are hats rather than headcount — one person may wear two, and the founder often wears the governor hat longest. The point of the table is not four hires; it is that each hat must be on a named head, because an unworn hat is a decision nobody is making.

Getting there without a reorg

The low-risk path is to run the new shape inside one bounded cycle before formalising anything. Pick a single loop — a 90-day growth loop is the natural container — assign the four hats for that loop only, and watch where decisions queue: that queue is your org chart telling you what it wants to become. A first evidence file lands quickly enough to make this concrete within days rather than quarters (here is what week one of AI research can deliver), and one honest cycle will teach more about your future team shape than any amount of organisational theory — this essay included.