An Enterprise Architect's Take on Salesforce in Claude
Salesforce and Anthropic have announced Claudeforce, and it's the most interesting thing to come out of the platform in a while.
Salesforce and Anthropic have announced Claudeforce, and it's the most interesting thing to come out of the platform in a while.
For starters, Salesforce and Anthropic. For those that don’t know, Salesforce has a $5B stake in Anthropic. By making this move, Anthropic becomes even more embedded as the AI backbone of the biggest CRM in the world, which makes that company more valuable, and as we have seen, had a major impact to the Salesforce stock price.
Salesforce wins on both fronts. In addition to creating a ton of value through this circular arrangement, it will also shut-up all the analysts that have been trying to make SaaSpocalypse happen (It’s not going to happen, Gretchen).
Next up, this is an unlock for sales teams, admins and change teams. This may not be a leap forward from what you could already do with Claude through the MCPs, but it does mean that a lot of work has been done for you with these pre-made sales skills (assuming they work better than the accelerators in the some of the versions of the Industries Clouds).
This is great for people that don’t have internal Claude skills, or people that are working with the partners that love to shout about their ‘Claude Practices’ before they actually learn anything about Claude, let alone become Claude partners.
It also makes a difference in how easy it is to roll-out. Creating one admin connect versus getting hundreds or thousands of users to go through an OAuth flow is a major win. This flip side is that this also moves from a per-user type pricing many people are getting from Anthropic to a usage pricing. More on this later…
Finally, this is a genuine additional step towards truly headless CRM, so in that sense this benefits everyone that wants to move into a more AI integrated world and that are tired of looking at the Salesforce UI.
While I appreciate that my tone may suggest I am not personally that excited about this announcement, this is the part that does excite me. The ability to leverage all the structure and automations that I have long loved about Salesforce while serving it where I actually work is great.
It won’t make sales teams better at data entry, but it should make it easier for them and that is the first step for good data and accurate forecasts.
Procurement
As mentioned earlier, this is another shift towards consumption based pricing, and it will not be available for purchase on a single piece of paper. This could become a procurement headache for both reasons. Pricing for this product has not been released, so it is too early to say if this will make things more or less expensive, but it will add pricing uncertainty along with additional paperwork. It is also worth noting that when Anthropic tested this on themselves, they saw usage of Salesforce grow 5x in a quarter. This is a great adoption metric and hopefully means they were realizing more value from Salesforce, but it also suggests that consumption spiked. AI usage and consumption is a much bigger subject though, and at ThirdEye we have been working on some cool ways to track consumption across platforms - so if this is of interest to you come talk to us!
API Limits
The current API thresholds were meant for human users and known integrations. It is a lot harder to predict what agents will do to API consumption. Again, not a blocker, but it would suggest that just because you CAN roll this out to all users at once doesn’t mean you SHOULD roll this out to all users at once. Start with a sample group and monitor what it is doing to your governor limits.
Governance
On the Governance side, it is mostly positive but there are aspects of the other side of the sword to be aware of. On one hand, it makes it much easier to keep centralized control over what functionality is allowed, and according to all documentation it keeps the Salesforce platform as the security layer to ensure that the right data is accessed by the right users. This is a huge win, though how exactly this works remains murky to me. We will have to design some tests for this to see how it actually is working. It will be a fun follow-up to this post. On the flipside, Claud tag in Slack appears to run as an integration user, so it will potentially obfuscate who is actually making changes via AI as it will all appear to be the AI user (similar to how automations can update records and just show the default automation user). There are also some things in the documentation that I can’t seem to find a clear answer on, but it would be worth those in regulated industries to ask before buying:
Does the Trust Boundary extend to Salesforce in Claude? (and if so, what i the contractual instrument that delivers zero data retention)
Can Bedrock inference for Agentforce be pinned to the UK or EU region?
Is this in scope for the Hyperforce EU Operating Zone?
Will there be notice before the model version changes, and can we pin a particular model version?
This is a brilliant move from a business perspective. It has some genuine gains that are worth some hype, but a lot of this is built on existing tooling. It also raises some questions and things to monitor.
-- Jeff Steinke
Get in touch and tell us about your project or challenge