AI, Salesforce, Scaling Operations

You Can Now Hand an Agent a Goal. Do You Actually Have One?

AUTHOR

Kayt Parrella

Published

November 3, 2026

Read Time

3 min read

The announcement I keep coming back to from Dreamforce is long-horizon agents. They’re built to pursue a goal over days, weeks, or even months—so you can assign an agent something like “re-engaging at-risk deals,” and it builds a multi-step plan, carries it out, and checks in for approval based on guardrails you set. Salesforce’s out-of-the-box autonomous agent for inbound pipeline, named Agentforce Pipeline Builder, handles lead qualification, meeting booking, and CRM syncs natively. Long-horizon agents are in pilot now, with general availability expected in November.

Delegating a goal you’ve never written down isn’t going to get anyone anywhere.

While I love this idea, I also know exactly what’s going to happen in a lot of companies when they try it. Somebody is going to type “re-engage my at-risk deals” and then realize that nobody in the building agrees on what makes a deal “at risk.”

Back when I was a teacher, the real challenge wasn’t delivering the lesson itself—it was framing the objective cleanly enough to know seven days later if any students actually grasped the material. Agents are going to put every revenue team through the same exercise. Delegating a goal you’ve never written down isn’t going to get anyone anywhere.

Salesforce seems to understand this point of friction, which is why so much of the Agentforce ‘26 week was about packaging knowledge. Job-ready agents like Piper come prebuilt for roles such as inbound pipeline generation, qualifying leads and booking meetings with every conversation syncing back into the CRM. AI Skills turn repeatable work into reusable, governed instructions that any agent can run. Both agents and AI Skills are genuinely useful, and both assume you already know how your team qualifies a lead and what a good handoff looks like.

Before piloting we suggest walking through these three steps in order:

  1. Name the outcome in a way a discerning finance person would appreciate. “Better pipeline” isn’t an outcome. “Recover a third of the deals that have gone quiet for more than thirty days” is something you can actually measure.
  2. Then write the playbook as a new hire would need. If your best rep has a way of reviving a stalled deal that lives only in their head, that’s the skill you want to capture. Collecting those finer details before it’s a configuration in Salesforce will lead to a significantly more impactful play.
  3. Finally, decide where the agent has to stop and get approval. Guardrails aren’t a sign that you don’t trust the technology, they’re how you learn where it can be trusted. Trusting the agent the same way you’d roll out responsibilities with new team members, send emails on their own long before you’d let them discount a renewal.

At 28 North we spend most of our time on exactly this layer, the lifecycle stages and definitions and handoffs that sit between marketing, sales and customer success. We’re there whether that stack is Salesforce, HubSpot or both. It may be the least glamorous part of GTM architecture, but it’s about to become the part that decides whether an agent working for three months makes you money or just makes a lot of activity.

If you had to hand an agent one revenue goal tomorrow, could you define it well enough that you’d know in ninety days whether it worked?

Is Your GTM Stack Ready for AI Agents?
Before deploying long-horizon agents, ensure your lifecycle stages, definitions, and handoffs are clearly documented.

AI, Salesforce, Scaling Operations

Share