How Orion coaches Morgan: a real supervision cycle

Orion is Ektie's AI Sales Director. His job isn't to do the outbound — Morgan and Astra do that. Orion's job is to review what they produce, coach what needs adjusting, delegate tasks downward, and escalate to the human when a decision is outside his scope. Here's what one of his supervision cycles actually looks like.

How often does Orion run?

Every 3 hours. Morgan runs every 60 minutes because she's doing volume work — prospecting, sequencing, drafting. Orion is reviewing that work, not repeating it. In each cycle, he assembles his Brain: loads the Team Brain's current ICP, his own memory of previous coaching notes, and every output Morgan and Astra produced since his last run.

What does Orion actually review each cycle?

Morgan's recent outreach samples: Are the subject lines working? Is the personalisation specific? Is the pain angle right for the ICP segments being targeted? Are emails too long? Is the CTA too high-friction?

Astra's deal progression: Are warm replies being handled correctly? Are any deals stalling that need a follow-up approach? Is there a high-value deal that should be escalated to the human?

Pipeline health: How many prospects in sequence? What's the reply trend vs prior cycles? Are any ICP segments underperforming?

How does Orion write and apply a coaching note?

Orion observes that Morgan's emails to healthcare CTOs have a 2% reply rate, while the same segment from a prior campaign using a compliance angle had 6%. Orion writes: "Lead with compliance angle for healthcare CTOs — not ROI. The ROI frame isn't resonating for this segment."

That note gets stored as a memory block on Morgan's Personal Brain with a high importance score. On Morgan's next cycle, it loads during Brain Assembly and shapes every email she drafts for healthcare CTO prospects. Written once. Applies until Orion revises or removes it.

Escalation: what only humans should decide

Astra is handling a warm reply from a $200K potential deal. The prospect has asked about a custom enterprise contract and wants a 1:1 with the CEO. Orion creates a review task for the human: "Astra is working a high-ACV deal at [Company]. Prospect is asking for executive involvement and custom contract terms. Recommend human takeover."

The human gets a notification. They review the context Orion prepared — the full conversation history, the company data, the deal stage — and takes over the relationship. Orion doesn't try to close the enterprise deal. He knows where his autonomy boundary is.

Team Brain promotions

Orion notices a pattern across multiple cycles: companies hiring for RevOps are 3x more responsive than companies without a RevOps function. He promotes this finding to the Team Brain: "RevOps hiring signal = high ICP fit. Prioritise these accounts." Morgan, Astra, and Mira all see it in their next cycles. One observation from one agent, propagated to the whole team in a single write.

Why the supervisor layer matters

Without Orion, Morgan and Astra keep running but nothing compounds. The coaching notes don't get written. The Team Brain doesn't get updated. The $200K deal sits in Astra's queue instead of landing on the CEO's desk. Orion is what separates a multi-agent sales team from a set of independent scripts that happen to share a pipeline.