Most AI agents are good at reading and interpreting things. They can summarize the state of your pipeline, draft an email to a prospect, answer a question about something in CRM. But the real magic is when agents take the next step. When they actually route the lead, update the account record, move the deal forward. A human usually still has to go click a button.
We're starting to see that change, and one conversation with a fast-growing AI company made it concrete.
The setup
This company had just signed on with Gradient Works and needed to move fast. They had a rep training session days away, new AE and BDR hires starting, and a pile of accounts sitting in an ad hoc assignment system that had outgrown itself. The plan was what Gradient Works is good at: dynamic books of high-fit accounts, paired AE/BDR ownership, a workflow to claim and return accounts, and daily distributions driven by clearly defined target books.
None of that is unusual. What made the conversation interesting was what they wanted to build on top of it.
"Can our agent just do this?"
This company uses its own AI agent internally, with a robust library of skills it can use to take action in the tools the team relies on. Their question for us wasn't can we see our routing assignments via AI? (Yes you can though.) It was can our agent actually trigger the routing itself?
When a BDR sets a meeting with a new prospect, they should be able to quickly and easily create an opportunity for that account and route it to the next AE. That sounds simple, but there's a lot going on in that process:
-
Take a contact or lead, and make sure it's attached to the right account
-
Create a new opportunity on that account
-
Figure out which AE is up next (and make sure that AE is eligible for this specific account and has capacity to work this account right now)
-
Route that prospect to the AE, and alert the AE about their new opportunity
-
Give the BDR a new prospect in place of the one they just handed off
Not only is there a bunch of logic in each of those steps, but we also expect this BDR -> AE handoff process to happen instantly with very little effort on the part of the BDR or AE (or their manager!).
So that means the AI agent should be able to say "create and route an opportunity for this account," have that call the same routing logic a rep would trigger from a button in Salesforce, get the new rep assignment, and trigger the account ownership change. All without a person in the loop and ideally without even going into the Salesforce UI. A BDR should be able to just tell their agent "move this account", and it should do it.
That's how the Gradient Works MCP works. The MCP exposes the same routing, ownership, and target book logic that powers our UI as callable actions any agent can invoke. The user's agent doesn't need to know how our round robin works internally; it just needs to ask.
Two front doors, one path
One detail on how this works is worth calling out, because it's a design decision we think more teams will run into as agent-triggered actions become normal. We don't want the UI button and the MCP call to be two different implementations of the same logic. They're two front doors into the same backend flow.
That matters for a boring but important reason: latency. An agent call has to travel through more hops than a native button click, so for time-sensitive fairness logic, both paths need to exist and both need to hit the identical rules. Build the logic once, expose it twice.
Letting an agent look without letting it jump the line
The other question this team raised: could their agent check who's next in the round robin, without disrupting fairness for everyone else?
That's a reasonable worry, because a routing queue only stays fair if querying it doesn't change it. The answer is that Gradient Works only advances the queue on an actual assignment, never on a lookup. So an agent (or a rep) can ask "who's up next" as many times as it wants, and the queue holds steady until someone actually gets assigned. Read access doesn't cost you fairness.
What's next: agent-scored accounts
The same team is also thinking about where AI fits earlier in the pipeline, like using a model to score accounts and feed that ranking directly into target book prioritization, so reps spend their time claiming the accounts most worth claiming rather than working a list in arrival order. That will likely be the next step in this implementation: target books that don't just define who's eligible for an account, but rank them intelligently before a human or an agent ever picks one up.
The goal is to get the right accounts in the hands of the right reps at the right time, as effortlessly as possible so sellers only have to focus on what they do best: selling (and not CRM updates).
The bigger pattern
We keep hearing versions of the same request. GTM teams don't want to just let agents read our GTM systems, but want them to act on them, using the same rules a human would follow (and hopefully both faster and simpler). That's the whole premise of the Gradient Works MCP. Your agent doesn't have to be a second interface bolted onto ours. It can be a first-class way of operating Gradient Works, right alongside the UI.
If you're building your own AI agent or skill and want it to be able to trigger routing, account ownership changes, or query your setup in plain language, the Gradient Works MCP is available now (even on our free plans!). Get started →
