On 15 April Salesforce launched Headless 360 under a title that leaves no doubt: "No Browser Required".
The idea is simple. Every capability of the platform becomes an API, an MCP tool or a command-line command, so an agent like Claude Code or Cursor uses it without ever opening a page.
I have been doing this for months. From Claude, over MCP, I read an org’s metadata, query the data and apply changes with the Salesforce CLI. I no longer open the interface.
If nobody opens the interface any more, the job of whoever configures it changes.
What shrinks
Part of the admin’s work has always been translation: someone asks for a field, a view, a report, and the admin clicks until it exists. An agent already does that part, in minutes, from a single sentence. Anyone who defines themselves by their clicks has a problem.
What grows
Salesforce says it in the same announcement: permissions, sharing rules and controls stay in the platform, and agents work inside them. Someone has to draw those boundaries.
- who can see what, person or agent;
- which actions an agent takes alone and which need an approval;
- which rules really apply, exceptions included, written so a machine can read them;
- who changed what, when the writer was not a person.
It is the same job moved up a floor: from building the pages to governing the context.
The admin is not disappearing. They stop clicking and start deciding what whoever clicks is allowed to do, person or agent.
What I would do tomorrow, as an admin
I would install the Salesforce CLI and connect an agent to a sandbox. Not to learn a tool, but to see how much of my job an agent already does. I would spend the rest of the week on permission sets and sharing rules: that is where the work is now.
You do not need the CRM any more. And it has never mattered more.
In your org, who decides today what an agent is allowed to do?