Reading the queue from the intelligence service.
Fetching outstanding proposals, the agent catalogue, your approval permissions and the directory that names the people who decided. Nothing is approved or unapproved while this says so.
What the agents have recommended, and what a named human has decided about it. An agent proposes; it never acts. Nothing on this queue has taken effect until somebody other than the person it was generated for has approved it and released it.
Your session has expired. Sign in again.
This is the queue being unreachable, not the queue being empty. No conclusion about outstanding approvals can be drawn from this screen while it says this.
Open a proposal to read its evidence and decide it. Deciding from a list, without the shortfalls and the audience in front of you, is the failure the approval gate exists to prevent.
No proposals match this filter. Others may exist outside it.
Each agent states what it needs a human for, and which permission that decision requires. Every one of those permissions is human-only, so no model can hold it.
| Agent | Can propose | If approved | Decided with |
|---|---|---|---|
| Insights AgentExplains what changed in a cohort or a customer, with the counts the explanation rests on. | insight.findingA described change in behaviour or value, with the observation counts behind it. | AdvisoryRecords that a human accepted the finding. Nothing executes. | insight.action.approveAccepting a finding is what makes it actionable for the team. A finding nobody signed for is a rumour with a chart. |
| Insights AgentExplains what changed in a cohort or a customer, with the counts the explanation rests on. | insight.customer_noteA grounded note to write onto the customer record, citing the figures it came from. |
| customer.note.writeReleased to the owning workflow with the approval attached. |
ai.draft.approveA note on the customer record is read by agents handling that customer later and is disclosable to them. A model may draft it; a person owns it. |
| Decisioning AgentRecommends the next best action per customer, and never performs one — `nba.execute` is PROHIBITED for any AI actor. | decisioning.next_best_actionA ranked recommendation per customer, with the score and reason codes it rests on. | AdvisoryRecords that a human accepted the finding. Nothing executes. | insight.action.approveAdvisory by design. Executing a next best action is a separate human act under the existing policy gate; the AI boundary classifies `nba.execute` as PROHIBITED so no approval here can reach it. |
| Decisioning AgentRecommends the next best action per customer, and never performs one — `nba.execute` is PROHIBITED for any AI actor. | decisioning.soft_suppressionA recommendation to stop contacting a customer on a channel, with the signal that prompted it. | suppression.soft.applyReleased to the owning workflow with the approval attached. | customer.suppression.writeChanges whether a real person can be contacted. Only ever in the restrictive direction from an agent: removing a suppression is PROHIBITED. |
| Campaign AgentDrafts campaigns and asks a human to approve the send; it cannot send, and it cannot approve. | campaign.draftA campaign definition — audience, channel, objective, schedule. | AdvisoryRecords that a human accepted the finding. Nothing executes. | ai.draft.approveAccepting the draft moves it into the campaign workflow, where the existing approval and send gates still apply in full. |
| Campaign AgentDrafts campaigns and asks a human to approve the send; it cannot send, and it cannot approve. | campaign.sendA request to send an approved campaign to its audience. | campaign.sendReleased to the owning workflow with the approval attached. | campaign.sendAn outward-facing commitment to real customers. The costliest irreversible action in the product. |
| Campaign AgentDrafts campaigns and asks a human to approve the send; it cannot send, and it cannot approve. | campaign.journey_activationA request to put a drafted journey live. | journey.activateReleased to the owning workflow with the approval attached. | journey.activateAn active journey contacts customers on its own schedule from then on, without anyone pressing anything again. |
| Creative AgentWrites channel-appropriate copy variants for a human to choose between. | creative.variantCopy variants for one channel and locale, with the brand and compliance constraints applied. | AdvisoryRecords that a human accepted the finding. Nothing executes. | ai.draft.approveCopy that reaches a customer carries regulated wording. A person reads it before it is attached to anything. |
| Experimentation AgentDesigns experiments and reads results out; concluding one stays a human action in the experiments module. | experiment.designA variant split, primary metric and minimum sample size for a proposed test. | AdvisoryRecords that a human accepted the finding. Nothing executes. | ai.draft.approveA design a human has read is a design somebody can defend when the result is surprising. |
| Experimentation AgentDesigns experiments and reads results out; concluding one stays a human action in the experiments module. | experiment.readoutA readout of a running experiment, including whether it has reached its sample size. | AdvisoryRecords that a human accepted the finding. Nothing executes. | experiment.concludeAdvisory. Approving the readout records that a human accepted it; the experiment is concluded in the experiments module by that human, not here. |
| Compliance AgentChecks proposed activity against consent, responsible-gaming and jurisdictional rules, and recommends holds. | compliance.reviewA consent, responsible-gaming and jurisdictional review of a proposed campaign or audience. | AdvisoryRecords that a human accepted the finding. Nothing executes. | insight.action.approveA review a person has signed is the artefact a regulator asks for. An unsigned one is a draft. |
| Compliance AgentChecks proposed activity against consent, responsible-gaming and jurisdictional rules, and recommends holds. | compliance.holdA recommendation to hold a campaign, naming the rule it appears to breach. | AdvisoryRecords that a human accepted the finding. Nothing executes. | campaign.approveAdvisory in the restrictive direction only. This agent can recommend stopping something; it has no proposal kind that releases a hold, lifts a protection or clears a suppression, because those are PROHIBITED for an AI actor outright. |