Publishing on a cadence
The most common first project. A researcher agent gathers the material, a writer drafts, an editor checks it against your rules and voice, an image agent produces the artwork, and a publisher pushes it live on the schedule you set.
What you still decide
The topics worth covering, the voice, and — until you trust it — whether each piece goes out. Approve from the Publisher feed, or let a project publish straight through once it has earned it.
Where it catches people out
Cadence is a promise. Setting a project to publish daily when you only have enough to say weekly produces thin work on schedule, which is worse than nothing. Start slower than you think you need.
Working the inbox
Agents read what lands, sort it, draft the replies that are obvious, and put anything needing a human decision in front of you with the context already gathered — so you are deciding, not digging.
- Triage as it arrives, rather than in a batch when you finally sit down.
- Drafted replies waiting for you to glance at and send.
- Escalation with context — the thread, the history, and what it thinks you should know.
Where it catches people out
Cost follows mail volume, not a schedule you control. A busy inbox is the job most likely to push you up an engine tier, so watch it for a fortnight before assuming the cheap tier holds.
Messaging and approvals
Connect Telegram and the whole loop fits in your pocket. Agents report in, ask when they are unsure, and take instruction — so approving work does not mean opening a laptop.
This is the one that changes how much you actually use the product. Approval friction is what kills automation: if saying yes takes three clicks and a login, work queues up and the cadence dies.
Watching and reporting
A standing job that checks something on a loop — a source, a competitor, a feed — and tells you only when something meaningful changed. The value is in the silence: no report means nothing happened, which is information too.
Where it catches people out
This is the job that quietly eats an engine tier, because unlike a publishing cadence it never stops. Set the loop as slack as you can tolerate.
Running several at once
Each project keeps its own memory, rules, credentials and voice, so running five brands in one workspace does not blur them together. Shared engines, isolated context.
You can also point different engines at different projects — something cheap on the internal grind, something stronger on the customer-facing work. The bring-your-own-AI guide covers how that affects the bill.
What it is not for
Work you cannot check. If there is no way for you — or another agent — to tell whether the output is right, automating it just produces confident nonsense faster.
Anything where being wrong once is unacceptable. Agents are good; they are not infallible, and the platform is built on the assumption that a human reviews what matters. If a job has no tolerance for a bad draft ever reaching daylight, keep a person in the loop permanently rather than graduating it to autonomous.
Got a job in mind that is not on this list?
Describe it and we will tell you honestly whether it is a good fit, what it would take, and what it would cost to run.