Most sales teams we talk to in Sydney and Melbourne have the same complaint about their CRM: the data is only as good as the rep's willingness to type it in on a Friday afternoon. In September 2026 Anthropic and Salesforce put a beta plugin called Salesforce in Claude into the hands of paid Claude users, and it goes straight at that problem. It brings a seller's accounts, opportunities and pipeline into Claude, under the permissions Salesforce already enforces, and lets Claude draft the updates a rep would otherwise skip.
This post covers what the plugin does, how its governance model works, and what an Australian sales leader should check before switching it on.
What is Salesforce in Claude?
Salesforce in Claude is a beta plugin that connects Claude to a seller's Salesforce data and Slack workspace through two connectors, then adds 37 skills for account research, call prep, pipeline review and CRM updates. Salesforce remains the system of record, reps sign in with their existing Salesforce credentials, and Claude can only see the records those credentials already allow.
It is available in beta on all paid Claude plans through Salesforce AgentExchange, and an administrator provisions it for the whole org rather than each rep installing it alone. According to the Claude announcement, it is already live at GitLab, Siemens and Legora, with 7,000 Salesforce sellers using it.
The five jobs it takes off a rep's plate
The 37 skills group into five practical workflows. Each one ends with Claude proposing a change and the seller deciding whether it goes into Salesforce.
Daily brief: today's meetings, deals closing soon, at-risk opportunities and unread threads that need a reply, with suggested stage, close-date and task updates.
Call prep: context pulled from Salesforce, Slack and email, with stakeholders who are missing from the account added as contacts.
Deal scoring: opportunities scored against the team's own sales methodology, plus a drafted business case and mutual close plan.
Post-call follow-up: transcripts and notes turned into a follow-up email, a Slack deal-channel summary and a draft opportunity update.
Pipeline dashboard: coverage by stage, at-risk deals and drill-down by account, with a forecast narrative written for leadership.
None of these are new ideas. What changes is where they happen. Instead of a rep switching between the CRM, Slack, the inbox and a notes app, the work sits in one Claude conversation that already knows the account.
How does the approval model keep CRM data safe?
Every change Claude proposes to Salesforce needs the seller's approval by default, so nothing is written to an opportunity, contact or task without a person clicking yes. Combined with Salesforce's own permission model and the default of no training on customer data for Team and Enterprise plans, this gives compliance teams a clear line: Claude reads what the rep can read, and writes only what the rep confirms.
That matters in Australia because customer contact records are personal information under the Privacy Act 1988. An agent that edits CRM records on its own is hard to defend in a privacy impact assessment. One that drafts changes for a named, authenticated person to approve fits the access controls most organisations already document.
| Question | Default answer | What to confirm locally |
|---|---|---|
| Who can see what? | Existing Salesforce permissions | Profile and sharing rules are current |
| Can Claude change records? | Only after seller approval | Whether any auto-approve settings are enabled |
| Is our data used for training? | No, on Team and Enterprise | Which Claude plan each seller is on |
| Who switches it on? | An admin, org-wide | Which teams join the pilot first |
| What else is connected? | Salesforce and Slack | Slack channel access for deal rooms |
Where it fits against Agentforce and a custom build
Salesforce already sells its own agent layer. We compared the two approaches in Claude vs Salesforce Agentforce. The plugin narrows the gap: Claude now works inside Salesforce data directly, while still reaching Slack and email in the same conversation. For teams whose sellers already live in Claude for writing and research, that is less change to manage than adding a second assistant.
It also changes the maths on custom builds. Twelve months ago, getting Claude to read pipeline data meant an integration project. For the common seller workflows, that project is now a configuration exercise. Custom work still makes sense for unusual objects, heavy Apex logic or data held outside Salesforce, but it is no longer the starting point.
What should an Australian sales team check before turning it on?
The plugin is quick to enable, which is exactly why a short readiness check pays off. The biggest risk is not the software. It is switching on a tool that faithfully reflects messy permissions, stale opportunities and a sales methodology nobody has written down.
Permissions and data hygiene
Claude inherits whatever access a rep has. If your sharing rules are loose, the daily brief will surface records a rep should not be reading. Audit profiles first, and archive dead opportunities so the pipeline dashboard reflects reality.
Your methodology, written down
Deal scoring works against the team's own methodology. If MEDDICC or your in-house equivalent lives only in the head of your best manager, capture it before the pilot. Our guide to Claude Skills for sales teams covers how to turn those deal motions into something Claude can apply consistently.
A measured pilot
Start with one team of five to ten sellers for four to six weeks. Track time spent on CRM admin before and after, the share of opportunities with a next step and close date that are current, and forecast accuracy at the end of the period. As an illustrative example, if ten reps each recover three hours a week at a loaded cost of $95 an hour, that is roughly $148,000 a year in selling time. Plug your own numbers into our ROI calculator before you commit to a wider rollout.
What not to conclude from the launch
A beta with 7,000 sellers is a strong signal, but it is not proof the plugin suits every sales motion. The named customers are large, well-run Salesforce orgs. Teams with heavy customisation, multiple Salesforce instances or strict data residency rules should test those edge cases rather than assume they work. It is also worth remembering that approval by default only protects you if reps actually read what they approve, so build review into the pilot rather than trusting the button.
The plugin pairs naturally with the broader rollout questions we covered in AI-native revenue teams: who owns the skills, how usage is measured, and how managers coach with the new data.
Getting started
If your team runs on Salesforce and already uses Claude, this is one of the lowest-effort wins available right now. The work is in the preparation: clean permissions, a written methodology and a pilot with real measures. We help Australian organisations plan and run exactly that kind of rollout. See our services or book a conversation to talk through where it fits in your sales stack.
