Open your lead-assignment logic and read it the way a stranger would. Round-robin across six SDRs. Named accounts go to their AE. Enterprise is split by headcount and region. Nothing assigned after 6 p.m. goes out until morning. A partner-sourced deal stays with the partner manager for 30 days.

Every one of those rules is a compromise with a human constraint: capacity, time zones, comp plans, fairness, the fact that people sleep. Now an agent is answering inbound, researching accounts, and drafting follow-ups around the clock, and it has none of those constraints. So which of your rules still apply to it?

Routing was never really about speed

The usual story says routing exists to get the lead to someone fast. That's the smallest part of it. Most routing logic answers three other questions:

• Who owns this relationship? Named accounts, territories, and partner registrations decide who gets to talk to whom.

• Who gets credit? Assignment feeds comp, quota, and the Monday pipeline review. Change who touches a lead and you change who gets paid.

• What's fair? Round-robin and capacity caps exist so the rep with the best timing doesn't win every inbound.

None of these are about latency. They're rules of engagement. And almost all of them live somewhere an agent can't read: an assignment rule in the CRM, a territory spreadsheet, a comp plan PDF, and the memory of the RevOps lead who wrote them.

What breaks when the responder never sleeps

An agent that answers at 2 a.m. doesn't wait for the round-robin. It acts, and the rules it can't see get broken quietly:

• It replies to a new contact at a named account in the middle of an active opportunity, and the AE finds out from the buyer.

• It books a meeting for a lead whose rep is on leave, because the out-of-office lives in a calendar the agent was never connected to.

• It re-engages a partner-registered deal inside the protection window, and now you have a channel conflict.

• Two agents, one on inbound and one on outbound research, touch the same account in the same week with different messages.

None of these are model failures. The agent did what it was asked. It simply had no access to the rules of engagement, because those rules were written for humans, who absorb them through onboarding, Slack threads, and getting it wrong once in front of their manager.

This gets more urgent every quarter. Gartner found that 67% of B2B buyers prefer a rep-free buying experience, and it predicts that by 2028, 60% of brands will use agentic AI for one-to-one buyer interactions. The first touch is moving to software. The rules about who is allowed to make that touch haven't moved with it.

Routing rules are becoming governance

Here's the shift worth planning for: routing stops being a queue and becomes a policy. In a human team, routing hands the lead to a person and trusts that person to know the unwritten rules. In an agentic team there may be no hand-off at all, so the rules have to travel with every action.

That changes what good routing design looks like:

1. Ownership is a fact, not a field. "Who owns this account" has to be answerable by every agent before it acts, with context: open opportunities, active partner registrations, and recent touches by anyone, human or agent.

2. Rules are written once and enforced everywhere. If your inbound agent, your research agent, and the one a rep plugged in all carry their own copy of the territory logic, you'll have three versions by next quarter. Salesforce found that 51% of sales leaders say tech silos delay or limit their AI initiatives. Routing logic scattered across tools is that silo in miniature.

3. Some actions route to a person, not a queue. Replying inside a named account, touching an open deal, anything inside a partner protection window: those should land in an approval queue for the owner rather than go out automatically.

4. Credit follows the outcome. When an agent sources or advances a deal, the attribution rules should say where credit goes before the comp dispute, not after it.

What RevOps can do this quarter

You don't need new software to start. You need the rules written down in a form a machine can use.

• Inventory the unwritten rules. Ask your AEs and partner managers what an agent should never do in their book. Their answers are your real routing policy.

• Separate "who responds" from "who owns." Agents can respond fast. Ownership and credit should stay explicit and set by people.

• Decide what gets gated. List the actions where an agent must ask the owner first. Start strict and loosen as trust is earned.

• Keep one source of truth. Territory, ownership, and protection windows belong in one place every agent reads, not pasted into each agent's prompt.

The layer underneath

This is the problem we're building wysdym around. wysdym is the operating layer for agentic GTM. It grounds every agent in your GTM truth, runs the motion under your governance, and gets sharper with every deal you close. The context an agent needs before it acts, like who owns the account, what's open, and who touched it last, belongs in Cortex, the shared memory every agent reads from. Governance applies per-agent permissions and approval queues to every write, so an agent reaching into a named account asks the owner first. And because every agent connects through the Gateway, one MCP door to your stack, the rules are defined once instead of copied into every tool. Here's how the pillars fit together.

Your routing rules were a good answer to a human problem. The next version has to be readable by software that never sleeps, and enforced on every action, whichever agent takes it.

We're building this with a small group of design partners: RevOps and sales leaders already running more than one agent against the same book of business. If your rules of engagement live in a spreadsheet and someone's head, we'd like to compare notes.