A campaign is a plan; an agent is a worker. What happens where the two meet — which agent runs your campaign, why a campaign has no autonomy setting, and why campaign drafts always wait for you.
Jump to module
The cleanest way to hold the difference:
An agent is a worker you hire. A campaign is a brief you hand it.
An agent has a name, a personality, a role, jobs, a schedule and an autonomy level. It exists indefinitely. It is who does the work.
A campaign has a subject, an angle, a window and a plan. It has no personality, no schedule of its own, and no autonomy. It is what the work is about.
They are not alternatives. A campaign runs through an agent — every campaign week is executed by a content-creator agent in your workspace. If you have never hired one, the campaign hires one for you.
| Agent | Campaign | |
|---|---|---|
| What it is | A worker | A brief |
| Lifespan | Indefinite | 1–4 weeks |
| Has a personality | Yes | No |
| Has an autonomy level | Yes | No — always review |
| Has a subject | No | Yes, a theme |
| Has an editorial angle | Implied by its persona | Explicit, one of six |
| Can publish without you | Yes, if set to auto | Never |
| Judged by | Ongoing output | A single readable outcome |
You do not assign a campaign to an agent, and there is no picker for it. The run resolves the right agent itself, in this order:
Two consequences worth knowing:
If your campaign will not start and the message names a paused agent, resume that agent. This is a common and completely silent-looking failure otherwise — the campaign is fine, the plan is fine, and nothing will happen.
Hiring an agent for a campaign does not switch on recurring posting. When an agent is auto-hired this way, the recurring job that normally comes with a new hire is removed immediately. Starting a campaign must never quietly turn on a standing schedule you did not ask for.
Each workflow the platform runs has a hidden system job behind it, shared by every run of that kind. It carries none of your campaign's per-run configuration — no accounts, no plan. That is deliberate: the row is shared, so writing your campaign's targeting onto it would put two concurrently running campaigns in a race over the same field.
Your campaign's own settings ride on the run itself. This is why two campaigns with different accounts, different voices and different tones can run in the same week without interfering.
Agents have four autonomy levels:
| Level | Behaviour |
|---|---|
| Manual | Never acts on its own |
| Assisted | Drafts, then waits for approval |
| Auto | Approves and acts by itself |
| Aggressive | Like auto, and proposes new work |
Campaigns have none of these, and the reason is a single design decision:
A campaign always requests review. Its drafts land in the Approve queue, every time, regardless of your workspace autonomy or the agent's level.
You pressed "activate a plan", not "post these". A plan that published itself the moment you saved it would make the activation button the most dangerous control in the product.
So a workspace set to fully autonomous will still have its campaign drafts wait for a person. If you want content that publishes without review, that is an agent with a job on a schedule — not a campaign.
A few related mechanics, since they interact:
Use an agent when:
Use a campaign when:
Use both, which is the normal case:
A content-creator agent gives you baseline presence. Campaigns sit on top for launches, promotions and pushes. An Engager agent handles the comments those campaigns generate. An Analyst tells you which angle earned attention, which is the input to your next campaign's angle.
"I'll make my agent do the campaign."
You can tell an agent to write about your launch, and it will. What you lose is everything a campaign gives you: the countable plan, the window, the single angle applied consistently across every piece, the derived result at the end, and the ability to duplicate the whole shape when it works. An agent given a subject produces posts. A campaign produces an answer to "did this angle work?"