Somewhere in your company there’s a sales playbook explaining how deals are supposed to happen. Somewhere else there’s a salesperson who actually knows how they happen. The gap between those two documents is where a lot of your revenue quietly goes to die.
Which is what makes Salesforce’s new sales agent, Hunter, worth a real look. Another AI tool that sends emails does nothing for me — my inbox has survived enough innovation already. But a system that carries context forward and adjusts as a deal shifts underneath it? Fine. You have my attention.
Salesforce says Hunter learns from CRM records, customer conversations, enablement material, seller activity, and approved outside signals, and that it works across prospecting, outreach, opportunity progression, proposals, and forecasting, with permissions and approvals built in. Those are Salesforce’s claims. The page doesn’t tell you how well they hold up in your pipeline, and neither do I — not yet.
The bigger story here isn’t sales productivity. It’s that we’re handing software the job of acting on a company’s commercial assumptions. Which means those assumptions had better be good, because nothing is left to catch them if they aren’t.
For years, a mediocre GTM strategy could hide behind the friction of actually executing it. A rep read between the lines of bad instructions. Someone noticed the prospect getting drip-emailed had already signed a contract and quietly killed the sequence. Human judgment was doing unpaid consulting work the whole time, patching over strategic confusion nobody wanted to name.
Take more of that judgment out of the loop, and “generate more pipeline” stops being a strategy and starts being a search term. It doesn’t answer who deserves outreach, what counts as real buying intent, or when to walk away. It just hands the confusion a target and a deadline.
Picture an account where three people have shown up in your data: one sat through a webinar, one downloaded an implementation guide, one keeps asking pointed questions about integration. A sales call from six months back also revealed the company can’t move until its current contract nears renewal. Every one of those signals is real. None of them means the same thing on its own.
If renewal is eighteen months out, a full meeting cadence probably burns goodwill for nothing. If renewal is close and someone’s now digging into implementation, that’s a real conversation waiting to happen. And if the integration questions are surfacing a genuine product gap, the right move might be to say so out loud. A system that treats all three signals as equally urgent hasn’t learned your business — it’s just automated the old habit of chasing every flicker of interest like it’s a hand raised in class.
This is exactly where marketing gets pulled into the blast radius. Once an agent is acting on buyer context, the quality of what marketing hands it stops being a brand question and becomes an operational one.
Positioning decides which problems you’re credible enough to touch. Research explains why those problems are urgent now instead of eventually. Content gives a buyer something to actually use at the moment they’re stuck. Evidence gives them cover to defend the decision to their own boss. That’s the raw material an agent has to work with — no more, no less.
Now imagine asking that agent to help a buyer through a hard decision when the entire content library is thought-leadership posts, feature announcements, and case studies where everything went perfectly. It’ll have plenty of words. It still won’t have the answers that make buying difficult in the first place: what does this actually demand from our team, where does this product fall short, what tripped up the last customer who looked like us, what would make this the wrong time to buy? A glossy PDF that can’t survive a hard question in a real sales conversation stays a glossy PDF no matter how well the retrieval system finds it.
There’s a second trap sitting inside the appeal of “learn from your best sellers.” Salesforce says Hunter studies your strongest deals to reverse-engineer your sales motion. Sounds smart. Also raises the obvious question: how do you tell a repeatable practice from a lucky break?
Your top rep might just have the best territory, the oldest relationships, or a client list your newer hires will never touch. A message that shows up across your winning deals may have caused those wins. Or it may have simply been in the room when they happened. History doesn’t interpret itself.
Before I’d trust that kind of pattern, I’d want it checked against the losses too — and the stalled deals, and the customers who bought excited and then struggled once they actually used the thing. I’d want to know if the pattern holds across segments or only in the one where your best rep happens to work. And I’d want a tripwire for the moment a once-reliable approach quietly stops working, because organizations get very good at one thing: repeating last year’s advantage long after it’s expired.
That’s the actual job — not deploying the agent, but building the loop that reviews what happened and changes the next decision because of it.
If outreach keeps landing meetings in one segment but those meetings rarely turn into real opportunities, more outreach isn’t the fix. The targeting or the pitch is wrong, and sending it faster just produces more of the wrong meetings, faster.
If deals keep stalling at the exact moment implementation comes up, that’s not a sales-execution problem. That’s information the product and qualification process both need, and pretending otherwise just moves the stall to a later, more expensive stage.
If the customers won by one particular pitch churn fast, that finding has to travel back upstream before anyone scales the pitch that produced them. Somebody has to own that judgment call. An agent can execute tirelessly inside an organization that has made zero progress on actually learning anything — those are two entirely separate skills, and only one of them ships with the software.
If I were piloting Hunter, I’d start small and specific: a defined set of dormant opportunities with a documented, real reason to re-engage. Clear account criteria. Clear evidence requirements. Clear approval steps. Clear stop conditions — because knowing when to back off is half of what makes a good rep good, even though nobody’s ever put “declined to bother a prospect” on a dashboard.
Then I’d watch three things: did the conversations that resulted actually help anyone, did opportunities move, and did the team learn something that changed its next move. Time saved counts. So does time spent cleaning up after a mistake the agent made with confidence.
Hunter is a real step toward software executing more of the GTM process, and there’s a genuine amount of research, admin, and coordination overhead worth automating away. But the easier execution gets, the more precise leadership has to get about what’s actually worth executing.
So before you hand your playbook to an agent: make sure your team can say out loud why it works, where it breaks, and what evidence would change your mind. The agent will do the work. Your assumptions are coming along for the ride either way.