What happens to a sales organization when something else decides what a good sales conversation looks like?
That question has an increasingly concrete answer. Salesforce and Anthropic’s Claudeforce — a plugin that puts Salesforce inside Claude — ships with 37 prebuilt sales skills. Meeting prep. Deal health review. Pipeline review. Objection handling. Close planning. Their description of it is “an AI CRO for every seller.”
That’s not our characterization. It’s theirs, and it’s an accurate one.
This is a real advance, and the sellers who get it will feel the difference in the first week. Salesforce’s own CRO says her reps start the day with the pipeline review already done. Anyone who has watched a good seller lose a morning assembling account history before making a single call knows what that’s worth.
So the question isn’t whether this helps. It will.
The question is what it will be reasoning from.
Preparation Isn’t Support. It’s Where the Selling Gets Decided.
There’s a comfortable way to think about AI in selling that goes something like this: the AI handles the preparation, the human handles the conversation. Division of labor. Everyone keeps the part they’re good at.
It’s a reassuring frame. It’s also wrong, in a way that matters.
Preparation isn’t a chore that happens before the selling starts. Preparation is where most of the selling gets decided. The questions you’ll ask, the person you’ll ask them of, how you’ll frame what this opportunity is really about, what you’ve concluded is standing in the way — all of it is settled before anyone joins the call. The conversation itself is largely the execution of decisions already made.
So when an AI does the prep, it hasn’t taken a task off your seller’s plate.
It has made those decisions instead of them.
The Risk Isn’t Bad Output. It’s Confident Output.
Here’s what makes this different from the tools that came before it.
A dashboard that’s wrong looks wrong. A report with a hole in it shows you the hole.
An AI-generated call plan never looks wrong. It arrives complete, organized, and articulate — whether it was built on a rigorous methodology or assembled from whatever patterns happened to be in the data. There is no visible difference between the two, because the output carries no signal about what produced it.
Which means telling them apart isn’t a matter of reading more carefully. It takes judgment about what a good plan should contain in the first place — and that judgment is either something your organization has built into its people, or it isn’t.
A plan that’s obviously bad gets ignored. A plan that’s confidently, plausibly, fluently wrong gets executed.
That’s confident failure, and it’s harder to catch than ordinary failure, because everything about it looks like success.
They Named the Problem Correctly. Then Solved Half of It.
Salesforce isn’t naive about this. Their own page puts it better than most: “A model can reason brilliantly and still be a genius who’s never seen your deals.”
That’s exactly right. And they’ve solved it — for data. Point Claude at your pipeline, your accounts, your history and your permissions, and it reaches what Salesforce calls the deterministic truth of your business.
But knowing the state of a deal and knowing what should happen next in it are different problems. Grounding an AI in your data tells it what’s recorded. It doesn’t tell it what’s good.
Which raises the question of how much is really in that record. The skills are described as “grounded in 27 years of Salesforce experience” — 27 years of knowing what sales organizations put into a CRM. A thinner thing than knowing how selling works.
How much thinner is not a matter of opinion. Validity surveyed 602 CRM users and administrators across the US, UK and Australia in 2025. Seventy-six percent said less than half their CRM data is accurate and complete. Thirty-seven percent said staff regularly fabricate data to tell leaders what they want to hear.
So the question actually worth answering is this one: how accurate a record is your CRM of the conversations your people are having with buyers?
There’s a qualifier in Salesforce’s own description worth noticing, too. The plan is drawn from every call, email and stage change on the record. That isn’t a claim to have the conversations. It’s a claim about whatever got logged — and the business issue the buyer described, the criteria they’ll really decide on, what got said in the room and what it meant, mostly never got logged anywhere.
Even where a call was captured, a transcript records the conversation that occurred. It can tell you what a seller asked. It cannot tell you what they failed to ask, because a question never asked leaves no trace.
The record shows activity, and it shows outcomes. What it can’t show is the quality of the thinking in between. That’s what a methodology governs, and no volume of CRM history will teach it.
And there’s a tell.
One of the Skills Is a Practice We’d Tell You to Stop
Among the 37 is win-loss-review.
We’ve argued before that loss reviews are largely a waste of time — buyers can’t reliably tell you why they didn’t buy, post-decision rationalization is well documented in behavioral psychology, and if a buyer can articulate their criteria after the decision, the real question is why nobody surfaced it while it still mattered.
Maybe someone weighed that and shipped it anyway, because customers ask for loss reviews and a plugin should cover what customers ask for. Maybe it’s there simply because loss reviews are what sales organizations do, so that’s what’s in the data.
We don’t know which. Neither do you — and that’s the problem in miniature. The skill arrives with no stated position on whether the practice is worth doing at all.
A capability built from how the industry behaves will faithfully reproduce the industry’s habits, including the ones a real methodology would tell you to drop. That isn’t a criticism of the engineering. It’s what grounding in practice instead of in principle produces.
Information Raises Activity. Only Practice Raises Proficiency.
Every team in every sport has film. Access to film has never been the differentiator. Two teams watch identical footage and one of them sees things the other doesn’t — because one has a system telling them what matters, plus the reps and the coaching to act on it. The other has a screen.
Selling is no different. Activity × Proficiency = Sales. Information tools — dashboards, content libraries, CRM, and now generated call plans — raise activity, sometimes dramatically. What they don’t touch is proficiency, because proficiency isn’t information. It’s judgment, and judgment gets built the way the team with the system builds it: repetition, coaching, and feedback on real attempts.
Tools can absolutely raise proficiency. They just have to be built for it, and most aren’t.
Which creates a second effect worth following. When every seller can generate a polished call plan in seconds, the plan stops being the differentiator. What’s left is whether the seller can tell a good plan from a bad one — and execute it in the room, with a real buyer, under pressure.
That’s trained, practiced and measured. It isn’t downloaded.
Defining the Standard Is the Work
None of this argues for keeping AI out of your sales process. That ship has sailed, and it should have.
It argues for deciding — deliberately, and in advance — what your organization means by a good discovery conversation, a qualified opportunity, a healthy deal. Not aspirationally. Specifically enough that two different managers would recognize the same thing in the same call, and specifically enough that it can be handed to a machine.
Because it is going to be handed to a machine. The only question is whose standard goes with it.
Salesforce summarizes the architecture as “Humans direct. Agents execute. Salesforce governs.” It’s a fair summary, and something is missing from it. Nothing in that sentence teaches anyone anything.
That part was never theirs to solve.
The Question to Put to Your Team
If we handed our AI the job of preparing every customer conversation starting Monday, what exactly would it be working from?
If the honest answer is “whatever it thinks good selling looks like,” you haven’t automated your methodology.
You’ve replaced it.
At Axiom Sales Kinetics we’ve spent thirty years helping sales teams coach, learn, and sell more effectively. If you’re working out what your standard actually is, start with the sales operating system underneath it — or visit us at www.axiomsaleskinetics.com.
