The gap between pilot and open beta is a few weeks
Salesforce says Salesforce in Claude is available to select pilot customers now and that it expects to launch in open beta in September 2026. Read those two clauses together. A pilot is something Salesforce chose you for, with a named contact on their side and a handful of sellers on yours. An open beta is something a seller can ask for on their own, and the release gives an enterprise team a few weeks to work out the difference.
We have watched this transition go badly on other platforms, and the product is rarely the reason. The reason is that the platform owner learns about the connection from a support ticket, after a rep has already been using it for a fortnight. Our read of the Claudeforce announcement is that the September date is the one to plan around, whether or not your org was in the pilot.
Who turns it on, and who finds out
The release describes Salesforce in Claude as a plugin. Sellers use it to reason over live revenue context, automate pipeline updates, and take governed action from Claude. What it does not say is who enables it once the beta is open. It could be a Salesforce admin approving a connected app, a Claude workspace admin approving a plugin, or the seller alone. That is the first question we would put to the account team, because the answer decides which console shows the connection first.
There are two admin consoles here and two teams who each assume they own it. The Salesforce admin sees a connected app and an OAuth grant. Whoever runs the Claude workspace sees a plugin and a policy. Neither sees the whole picture unless somebody decides they should. Our sibling piece on where authentication and permissions should be owned makes the case for settling that before the first grant rather than after it.
The skill list moves while you are testing it
The announcement counts 37 prebuilt sales skills in the plugin and names meeting prep, deal health review, and pipeline review as examples. It also says additional prebuilt skills will begin launching in Q3 2026, which overlaps the beta window. A pilot evaluated in August will not match the plugin a seller connects in late September. Your test results have a shelf life.
The useful split is between skills that read and skills that write. Meeting prep reads accounts, contacts, and activity history. A pipeline update writes to the opportunity, and a written close date or stage change lands on your validation rules, your approval process, and your forecast. Our piece on why connecting data is easier than governing actions goes deeper on that split. For the beta, we would keep a dated list of which skills are enabled and which of them write, and review it every time the plugin gains one.
Test the governance claim with your own rules
Salesforce says the plugin routes actions through Salesforce to help ensure business rules are always enforced when an action is taken. That is a claim about your org's rules, so it can only be tested in your org. Pick the three rules your sales operations lead cares most about. A stage that cannot move backward without a reason field. A close date that cannot land in a closed quarter. An amount change above a threshold that needs a manager. Then ask a pilot seller to make each of those changes from Claude and watch what the seller sees when the rule fires.
The exercise is about the error path, because the happy path will work in the demo. If a validation rule blocks the update, does the seller get the rule's own message, a generic failure, or a confirmation of something that never saved? Each of those produces a different support queue in October. Our guide on running a vendor demo on your terms covers how to script that with your own messy data.
Decide the stop conditions before September
An open beta will have a first incident. Somebody's pipeline gets updated in bulk, or a rep's session touches records nobody expected it to see, and the question in the room will be how to stop it in the next five minutes. The release does not describe a kill switch, so the team needs to know which ones it already has. Revoking the connected app in Salesforce cuts every seller off at once. Revoking one user's grant is quieter. Disabling the plugin on the Claude side is a third option that a different team owns.
Write down which of those you would reach for first, who has the rights to do it at six on a Friday evening, and what evidence gets kept so the incident can be explained on Monday. We have argued before that agents need an ordinary pause button, and a beta is exactly the period in which you find out whether yours works. Our agent readiness check is a longer version of the same conversation.
What we would put on the agenda
The release says nothing about pricing, licensing, or what an org has to do to qualify for the open beta, and we would rather flag the gap than guess. What it does give you is a month. The same questions apply to every agent surface that can write to CRM, and our Agentforce coverage keeps returning to them.
Take one question into the next Salesforce steering meeting. If a seller connects Claude to production on the first morning of the open beta, who in your organisation finds out, and how long does it take them?
