I was catching up on my reading last week and came across a piece on SalesforceBen which, I have to admit, made me laugh out loud: Salesforce is going to charge you when your AI agents talk to Salesforce. Not when they do something clever, mind you, but every single time an agent successfully calls the API or the MCP server. I read it twice, thought about it for a bit, and the more I thought about it the less funny it got, because I’ve seen this movie before. It’s called IBM.

What Salesforce Actually Announced

Here’s the gist (and bear in mind the details are still moving). Salesforce introduced what it calls Agentic Identity: any AI agent that wants to work with your Salesforce data has to be registered, gets its own credentials and permissions, and then every successful call it makes, through MCP or the plain old API, counts as a “Headless Platform Interaction”. Each one of those burns Flex Credits from your Digital Wallet (yes, that’s really what it’s called). How many credits per interaction? To be announced. When does the meter start? Also to be announced, with 30 days’ notice once the number makes it onto the rate card. For a sense of scale, Flex Credits sell at $500 per 100,000, and a standard Agentforce action costs 20 of them, so about 10 cents a go.

Now here’s the bit that matters: “traditional” integrations keep their current pricing. So the exact same API call, reading the exact same Account record, is free if a cron job somebody wrote in 2014 makes it, and metered if an agent makes it. Serving that call costs Salesforce exactly the same either way. The only difference is who (or rather what) is asking. Can you see what’s going on here?

The partner community certainly could. One ISV founder said Salesforce had “locked away three years of progress on the platform”, and the noise was loud enough that Salesforce opened talks with partners about it. Talks are good. I’d be very surprised if the meter goes away though.

Salesforce Is the New IBM

Let me take a detour through the mainframe, because I think it explains everything.

Plenty of banks and insurers still run the software their whole operation depends on (ledgers, payments, policies, the lot) on IBM mainframes. Every year they pay IBM a lot of money for the privilege, and for the big ones that means hundreds of millions of dollars. Every few years IBM comes up with a new pricing structure, or a new generation of “more performant” hardware with a higher price per unit (read “you’ll pay more for the same work”), and the bill goes up. Do the banks walk away? Of course not. Replacing decades of COBOL that moves real money around is complex, slow, and exactly the kind of risk that ends careers if it goes wrong. As such, they made their peace with it a long time ago. To the CFO, the IBM invoice is simply a line item in the yearly budget, somewhere next to rent and electricity. They’ve become immune to it.

Meanwhile, every startup and scaleup built in the last 15 years runs on distributed systems in AWS, Azure, GCP and the likes, and none of them would dream of buying a mainframe. IBM knows that too. It isn’t fighting for new customers on that front anymore; it prices for the ones who can’t leave.

Now swap “mainframe” for “Salesforce” and read that again.

Nobody New Is Signing Up

Here’s the thing: I don’t think Salesforce is winning many new logos among the AI-native companies, and I’d venture to say they know it. Ten years ago, if you were a growing company, you bought Salesforce because building your own CRM was a ridiculous idea. Nowadays, a team of 3 good engineers with the current AI coding tools can build a CRM that fits your sales process like a glove in a couple of months. It won’t have 20 years of features you’ll never use, it won’t need a certified admin, and it won’t need a consultancy to add a picklist value. I tinker with these tools every day (I’m still a software engineer at heart), and what a small team can ship nowadays would have sounded like science fiction not that long ago.

I know some of you will jump at my throat for this, so let me be fair: for a 5,000-person sales organisation with complex territories, approvals and compliance needs, that’s a much bigger job. But that’s precisely my point. Salesforce’s future customers look less and less like that, while its existing customers look exactly like that.

So if you can’t grow the club, you squeeze the members you’ve got. You put up list prices on the main editions (which Salesforce did again in 2025), you repackage, you invent credits and wallets so nobody can work out what anything costs, and, most importantly, you make it harder and harder to leave.

Why Agents Scare Salesforce

Let me explain why agents are the real threat here, and why the meter lands exactly on them.

Salesforce’s business model is, at its heart, a human logging in to a UI, priced per seat. AI agents are the first serious challenge to that. Once an agent can read and write your Salesforce data through MCP, your salesperson doesn’t need to open Salesforce anymore: they ask Claude, ChatGPT or your own internal agent, which pulls what it needs from Salesforce, your data warehouse, your email and your calendar, and gets on with it. Salesforce turns into a database sitting behind somebody else’s interface. A database with a very expensive per-seat price tag, I should add.

And once it’s “just a database”, the obvious question in every CTO’s head is “why are we paying this much for a database?”. Next thing you know, one workflow at a time moves out (the reporting first, then the forecasting, then the pipeline management), and there’s a lot less left inside Salesforce to charge for. Plugging agents in is step one of that slow unbundling, and the days of every SaaS product being a destination with its own UI are numbered. Salesforce knows all of this as well as the rest of us. It can’t stop it (it can hardly lock customers out of their own data), but it can put a toll booth on the road out. Charge per agent call and you make the “use any agent you like” route more expensive than the Agentforce route, and you get a shiny new revenue line to make up for the seats you’re going to lose. Bear in mind that since this summer Salesforce counts Headless 360 (the toolkit this metering is part of) and Slackbot in its Agentforce ARR, so your agents’ API calls will quite likely end up in an investor deck as “AI growth” 🙂

Now, don’t get me wrong, giving agents their own identity and permissions is a good idea, I’ll give them that. Running an agent under a human’s credentials is a security headache, and being able to see and limit exactly what each agent can touch is something I’d want in any system I’m responsible for. But you don’t need a meter to do identity. The identity is the security story; the meter is the business story. And it’s yet another desperate move from a dinosaur-era platform trying to slow down the inevitable.

Ok, granted, “dinosaur” sounds a bit harsh for a company heading for $46 billion of revenue this year. But IBM still makes a lot of money too, and nobody builds anything new on a mainframe. A platform can be very profitable and dying at the same time, as long as you measure “dying” by who’s still choosing to build on it.

Follow the Acquisitions

If you want a second opinion on where Salesforce itself thinks its future is, look at its numbers and at what it’s been buying. In the last quarter Salesforce reported $11.3 billion of revenue, up 11%, of which $456 million came from Informatica, which it bought last year. Take Informatica out and, by my maths on their own figures (the same quarter last year was $10.2 billion), the core business grew roughly 6%. The CFO talks about “second-half organic revenue reacceleration”, which is a polite way of saying it hasn’t happened yet.

Meanwhile the shopping list keeps growing: Slack for $27.7 billion, Informatica for about $8 billion, Contentful pending, and as of 20/Oct/2026 Braze for $5.66 billion to give its marketing business a kick. When a company’s growth story relies on the companies it buys rather than on the product its name is on, that’s telling. Even the analysts commenting on the Braze deal flagged bundled pricing that makes switching costlier at renewal. Go figure.

So What Now?

If you’re a CTO with Salesforce at the centre of your stack, here’s what I’d do on Monday morning:

• Find out where agent traffic is going to hit Salesforce: your own agents, your vendors’ agents, and the admins and partners already using Claude to configure the org. Put a rough number on it before Salesforce does.
• Make sure your customer data has a home you control (a warehouse kept in sync, at the very least), so Salesforce becomes one consumer of that data rather than the only place it lives.
• Stop building new workflows inside Salesforce. Every new Flow and custom object is another brick in the wall you’ll have to climb later.
• At renewal, ask for agent pricing in writing, with a cap. If they can’t tell you what an interaction costs, that tells you something too.

Bottom line: if you’re a Salesforce customer, know that you’re a member of a shrinking club, and a shrinking club gets squeezed harder every year to make up for the members who leave. Want to move with the times? Salesforce is not the answer. Or you can do what the banks did with IBM: make your peace with it, get used to a line item worth a few million dollars in your budget every year, and don’t change a thing. Just don’t act surprised when that line item grows.