Autonomous does not mean unaccountable.
Every draft is reviewed. Every action is logged. Every dial — approval mode, spend, and the switch itself — is in your hands, per brand and per channel. This is the page for the person who has to sign off on giving AI the keys.
Approval modes, per brand and per channel.
Start with everything requiring your sign-off. Loosen the leash only where you have earned the trust — never all at once, never everywhere by default.
Full human approval
Nothing schedules or sends without an explicit yes from a named person on your team. The safe default for every new brand and channel.
Reviewed batches
Approve a week's worth of drafts in one pass instead of one at a time — still your call, just less clicking.
Exception-only review
Low-risk, pre-approved content types post automatically; anything flagged by Observer or outside your guardrails still comes to you.
Supervised autopilot
The agents run the full loop for a channel you have proven out — you still see everything in the log and can pull the kill switch at any moment.
Modes are set per brand and per channel — a new product launch can stay on full approval while your evergreen newsletter runs on autopilot, on the same account, at the same time.
Observer checks every draft, before you ever see it.
Observer's whole job is to be the skeptic. Every piece from every agent — draft, image, email, landing page — is pressure-tested against your brand voice, your compliance rules and a risk pass before it reaches an approval queue.
- Brand voice — tone, terminology and style checked against what you set up.
- Claims & compliance — flags language your legal or brand team would flag.
- Risk & sensitivity — a pass for tone-deaf timing or sensitive topics.
A failed check does not get quietly dropped — it comes back to the responsible agent with the reason, and to you if it needs a human call.
What "governed" actually gives you.
Four controls that make an autonomous system something you can defend to your CFO, your legal team and your board.
Tamper-evident audit log
Every agent action, every human approval or rejection, every schedule and every send is timestamped and attributed — an immutable record you can hand to a compliance review as-is.
Cost ledger
Itemised spend by brand, agent and action, with caps you set — no surprise bill because a Discover run or an image batch got away from you.
Kill switches
Stop one agent, one channel, one brand, or the whole account in a single click. Nothing new drafts, schedules or sends until you flip it back on.
Brand-voice guardrails
The rules Writer drafts against and Observer checks for — set once per brand, enforced on every single piece, not left to a prompt someone forgot to reuse.
Miko here. I get asked "what if the AI just posts something bad" more than any other question. This page is the honest answer: it can't, quietly. Observer checks it, the log records it, and you hold the switch. Autonomy earns trust one governed action at a time.
Questions about governance.
Bring your compliance checklist. We'll go through it live.
Approval modes, the audit log, the cost ledger and the kill switch — shown, not just described.