Salesforce Just Said the Quiet Part: The Interface Is Not the Product Any More
Salesforce shipped a product this week whose main selling point is that you don't have to open Salesforce.
It's called AIforce, and it's an interface layer that pushes Salesforce data, workflows, business logic, permissions and governance out into whatever tool the person is already using. The first pieces: Claudeforce, a prebuilt server that puts Salesforce inside Claude with 37 ready-made sales skills, plus a plugin for Claude Code with 40 more. Slackforce, which lets you create accounts and update records inside a Slack conversation. And Agentforce Coworker, which Salesforce says reached 100,000 activated users in its first 35 days.
Marc Benioff's framing was "AI is creating an interface revolution". A trade headline put it less diplomatically: AI replaces the UI.
Read that again with your business owner's hat on. The company that spent twenty-five years making a screen your team lives in all day just shipped a product built on the premise that the screen is optional.
The interface was the moat
Think about what an enterprise application has actually been selling you.
Not storage. Not even the logic, mostly. It sold you a place where work happens, and everything valuable flowed from that. Seats are priced per person staring at the screen. Switching costs come from the months your team spent learning where the buttons are. Whole careers are built on knowing one vendor's interface well enough to be certified in it. An ecosystem of consultants exists to arrange those screens for you.
All of that sits on one assumption: that a human opens the app and does the work by hand.
Take the human out of the clicking and the assumption stops holding. If an agent does the pipeline hygiene, it doesn't care how pretty the dashboard is or where the button sits. It cares about three things: can I read what I need, am I allowed to do this, and will the system tell me whether it worked.
That's a very different product. And the interesting thing is that Salesforce, which had more to lose from that shift than almost anyone, decided to lead it rather than defend the old thing. The bet is that the value was never the screen: it's the data, the rules, the permissions and the record of who did what. Those survive the interface. The dashboard doesn't.
What gets valuable instead
If your software is going to be used by something that never attended a training session, a few unglamorous things become the product.
Permissions become the interface. When a person does something silly, they usually notice. An agent acting on behalf of ten people needs the rules to be real, enforced in the system, not implied by which menus someone was shown. Every vendor moving this way is going to discover which of its "permissions" were really just hidden buttons.
Business logic has to be written down somewhere readable. The rules that live in your most experienced person's head (never discount above 15% without asking, always check the credit note before refunding) have to exist where an agent can find them. Most companies have never written them down. This is the actual project, and it's not a technical one.
The audit trail becomes the trust. You can't watch an agent work, so the only way to be comfortable is to be able to reconstruct what it did. Which conversation, which record, which change, at whose request.
Notice that none of this is about model quality. It's about whether your business is legible to something that reads rather than looks.
What it means for the work
The shift most people will feel first is boring and enormous: fewer tabs.
The skill that used to make someone valuable inside a company was app fluency. Knowing the system, knowing the workaround, being the person you ask when the report won't generate. That skill is depreciating. What replaces it is knowing what should happen: the rules, the exceptions, the judgement calls, and the ability to check an agent's work well enough to sign it.
That's a promotion disguised as a threat, and it's the same conclusion we reached writing about uniquely human abilities back in June. The person who knew where the buttons were becomes the person who decides what's correct.
There's a catch, though, and it's worth saying out loud because the vendors won't.
An interface layer is not a coworker
Claudeforce lives inside Claude. Slackforce lives inside Slack. Every major vendor will now build a version of this, which means the surfaces are all becoming agent-ready while the agent itself stays borrowed. You get a beautifully connected set of tools, called by an assistant that belongs to someone else, remembers this conversation and not last quarter's, and has no identity anyone can contact.
Tools being reachable is the easy half. The hard half is having a worker with continuity: something that knows your business, holds the context of what happened three months ago, has its own identity so a client can write to it directly, and can be held to an audit trail you own. Otherwise you've upgraded your tools and still have nobody to hand the job to.
That's the bet we're making from the other side. Our AI employees work where the people already are, in email, WhatsApp and Slack, rather than asking anyone to come and sit in our interface. Same conclusion as Salesforce, different starting point: if the work is delegated, the screen is a detail.
What to do about it before the vendors decide for you
Three things, none of which require buying anything this week:
- Write down the rules. Pick your most repetitive process and get the decision rules out of people's heads and into a document. You need this whether your agent comes from Salesforce, from us, or from a model you run yourself.
- Check which of your systems can actually be called. Some of your stack is already agent-ready and some of it is a web form with a login. That difference is about to matter more than the feature comparison you made when you bought it.
- Decide who reviews. Before agents start touching records, name the human who signs off on what, and what the threshold is. Doing it now is a ten-minute conversation. Doing it after an incident is a different conversation.
The interface era isn't over because dashboards stopped being useful. It's over because the most valuable work in most companies was never the clicking, and we've finally got something else willing to do that part.
Want to test the most advanced AI employees? Try it here: https://Geta.Team
