When Should You NOT Use AI Marketing Agents?
AI marketing agents are the wrong choice when a decision carries legal or reputational risk that needs a human sign-off, when your ICP and brand voice aren't defined yet, when nobody has time to review and approve the work, or when the job is a single one-off deliverable rather than a sustained motion. Every one of these limits is visible in how Orbitable itself is built: Autopilot only proposes a plan, a human approves it in the Dock, and credits meter capacity so work pauses instead of running unsupervised.
When the decision carries legal, financial, or reputational risk
Crisis comms, pricing changes, legal claims, layoff messaging, regulatory language: these need a human who owns the consequence, not a draft that ships on its own. Orbitable's agents draft and recommend, they don't publish blind. Autopilot proposes a weekly plan only. Approval happens in the Dock, before anything moves forward. That's a deliberate constraint, not a limitation to route around. If you want zero-touch publishing on anything that could expose the business to liability, no responsible agent system, including this one, is built to do that.
When your ICP and brand voice aren't defined yet
Every agent in the fleet shares one world model per customer: ICP, brand voice, products, competitors, and an uploaded knowledge base. That shared context is the entire point, work by one agent compounds for every other agent. But an undefined or wrong ICP compounds just as fast. Feed the system a vague sense of who you sell to, and fifty agents across ten squads will produce fluent, confident copy for the wrong buyer, at a pace no single copywriter could match. Fix the world model first. Turning on the fleet before that is done multiplies the mistake, not the output.
When nobody is available to review and approve
Autopilot proposes a weekly plan, one focus, three to five missions, each with a why-now, a carry-over list, and an explicit not-doing list. It proposes only. Nothing executes until a person opens the Dock, reviews the pins on a draft, and approves it into the Library. That works well when someone checks in weekly. It stops working the moment there's no one with the time or authority to do that review. An unattended queue of proposed missions isn't automation, it's a growing backlog nobody is clearing.
When the job is a one-off, not a sustained motion
The GTM plan runs across six phases and twenty-six agent steps. The 22 playbooks and nine schedule templates are built for repeating cadences: weekly content, ongoing outbound, recurring reporting, not a single landing page needed by Friday. If the job is genuinely one-off, a fleet of 50 agents orchestrated by the Dispatcher is more infrastructure than the task needs. This earns its cost when the same world model and the same playbooks keep getting reused, week after week, not when it's spun up for a single deliverable and shelved.
When credits and plan size don't match the workload
Credits meter the work agents do, reviewing never costs credits, but the allowance is finite. Founder gives 3,000 credits/mo, Team 15,000/mo, Agency 80,000/mo, and only Enterprise is uncapped. When credits run out mid-month, work pauses until a prepaid credit pack tops it up or the month resets. There's no overage billing. That's useful discipline against runaway spend, but it also means an undersized plan against a bursty workload, a launch week, a big campaign push, will stall exactly when you need momentum most. Size the plan to the volume you actually expect, not the volume you hope to average across the year.
Good fit vs bad fit, side by side
FAQ
Can AI agents publish content without a human approving it first?
No. Autopilot proposes a weekly plan only, with a focus, three to five missions, and a why-now for each. Approval happens in the Dock before anything moves forward, and internal team notes are never visible to a client reviewing there.
What happens when an AI marketing plan runs out of credits mid-month?
Work pauses. There's no overage billing, so you either top up with a prepaid credit pack or wait for the monthly reset before the fleet resumes work.
Do AI agents need a defined ICP to work well?
Yes. Every agent draws on one shared world model per customer, covering ICP, brand voice, products, competitors, and an uploaded knowledge base. An undefined or wrong ICP gets inherited by every agent at once, so the error compounds as fast as the value would have.
Is it worth using a 50-agent platform for a single one-off task?
Usually not. The GTM plan, the 22 playbooks, and the nine schedule templates are built around repeating cadences across six phases and twenty-six agent steps. A single landing page or press release doesn't need that structure behind it.
Which plan should I choose if I'm not sure how much work I'll need?
Start with Founder ($89/mo, 3,000 credits) or Team ($349/mo, 15,000 credits) and watch how quickly credits burn against your real cadence. Move to Agency ($1,999/mo, 80,000 credits) or a custom Enterprise plan once that volume is proven, not before.