TL;DR | The Highlights
- Claudeforce pairs Salesforce and Anthropic to run your CRM data and workflows inside Claude, with Claude now the default model for Agentforce, Prompt Templates, and Vibes.
- Strip the branding and very little of the underlying capability is new. Most of it has been possible for months, since Headless 360 and Salesforce’s MCP servers opened the door.
- What actually changed is the on-ramp. Prebuilt skill plugins replace the homebrew setup early adopters had to build by hand, and one central admin connection replaces per-user authorization while preserving each user’s permissions.
- That shift matters more than any single feature. When setup stops being hard, being first to install stops being an advantage.
- The plugin inherits your org. It acts through your existing permissions and data, so a messy revenue process just gets a faster way to act on messy data.
If you follow Salesforce news, you have already seen the headline. Alongside its Q2 earnings, Salesforce and Anthropic announced Claudeforce, an expanded partnership that puts Claude inside Salesforce and Salesforce inside Claude. The centrepiece is Salesforce in Claude, a plugin that ships with a library of prebuilt sales skills so a seller can review pipeline, prepare for meetings, and take governed action on live CRM data without opening Salesforce at all. Claude also becomes the default model for Agentforce, Prompt Templates, and Vibes, all of it wrapped in Salesforce’s governance.
It reads like a capability leap. It isn’t. And mistaking it for one will point your team at exactly the wrong work.
Almost None of the Capability Is New
Here is the uncomfortable version, and we can say it because we have been building this ourselves. Nearly everything Claudeforce lets a revenue team do, you could already do.
The moment Salesforce shipped Headless 360 and its MCP servers, the hard boundary between “your CRM” and “an AI agent that reasons over your CRM” came down. The data, the workflows, the ability to have a model read live revenue context and take a governed action through Salesforce’s rules: that has been technically available for months. Teams who knew what they were doing were already wiring Claude into their Salesforce orgs and getting real work out of it.
The catch was that wiring it up was a project. You had to know which servers to stand up, how to scope permissions, what a good skill actually looked like, and how to keep the whole thing governed. It worked, but it took genuine engineering time, and it stayed the preserve of the people comfortable doing that.
What Changed Is the On-Ramp, Not the Engine
So if the capability was already there, what did this announcement actually deliver? A dramatically shorter on-ramp.
Instead of homebrewing skills, you install a plugin. Salesforce in Claude comes with prebuilt sales skills (pipeline review, deal health, meeting prep) built to actually use Claude’s reasoning, not generic prompts stapled to an API. There are bundles aimed at sales users and bundles aimed at admins, so the person setting it up is not starting from a blank canvas.
The authorization story changed too, and this is the part admins will feel most. An admin connects the Salesforce environment to Claude once, centrally. Individual users no longer each walk through their own authorization flow, and yet every user still operates inside their own permissions and context. You get central setup without flattening your permission model, which is usually the trade that makes these rollouts painful.
There is a smaller quality-of-life addition too: the Claude web and desktop apps now include UI enhancements for visualizing your Salesforce data directly inside them.
None of this is new capability. It’s friction removed from things you could already do, not new things to do.
When Setup Stops Being the Moat, Process Becomes It
This is where it gets interesting for anyone running a revenue function, because a lower barrier to entry quietly changes where the advantage lives.
When standing this up took real engineering effort, the effort itself was a kind of moat. Being early, being technical, being willing to homebrew separated the teams getting value from AI in their CRM from the teams still talking about it. That moat just evaporated. If your competitor can install the same plugin in an afternoon, being first to install it is worth almost nothing.
So what is left to compete on? The unglamorous part. How clearly is your revenue process actually defined? How clean are your permissions and your data? Do you know which decisions are worth handing to an agent and which are not?
Because Salesforce in Claude acts through your existing Salesforce, governed by your existing rules, on your existing data. It inherits your org exactly as it is. Point a fast, capable agent at a tidy, well-governed revenue process and it compounds. Point the same agent at a messy one and you have simply bought a quicker way to act on bad data.
A plugin cannot install judgment about your own business. The setup is what got easier. The thinking did not.
The Barrier Dropped. The Premium Moved.
Claudeforce is a genuinely useful release. Removing the homebrew tax on capabilities this powerful is worth celebrating, and we will happily stop rebuilding the same scaffolding by hand.
But the story underneath the announcement is not “AI can now do more in Salesforce.” It is that the technical barrier to using it just fell, which raises the premium on everything that was never technical in the first place. The teams that win the next year will not be the ones that installed the plugin first. They will be the ones whose revenue process, permissions, and data were in good enough shape that a capable agent made them faster instead of just louder.
That work does not come in a plugin, and it is exactly the work we do. If you want a clear-eyed read on whether your Salesforce org is ready to be pointed at an agent, or what it would take to get there, let’s chat.
Let's chat!
