I’ll be straight about what this one is. Tuesday was the keynote, and what I have for you tonight is a take on the morning, not the rebuild notes this series promised. Those come when I’ve had a demo in front of me long enough to pick it apart.
Start with the thing I expected. I walked in assuming the AI story would be oversold, and I was right, though not in the way I’d guessed. It was confusing more than it was inflated. There was a lot of it, told from several directions, and the successes presented on stage didn’t feel like adoption and results had been linked. A customer rolled something out. Sure. Voice AI at scale is cool, but did it drive revenue? If you have to sign off on a rollout, that link is the whole question, and I left without it.
Now the thing I’d want to rebuild. The demo that grabbed me was the plugin install for Claudeforce, the “Salesforce in Claude” piece: Claude, Anthropic’s assistant, gets a plugin that opens a governed view into a Salesforce org, and the plugin ships with a set of skills, the packaged tasks it already knows how to run. For the orgs I work in, that’s real visibility into the org itself, and it fits where those teams are trying to get value right now. So the install is the easy part. There’s not a lot to build.
What I don’t know is how to turn it on safely. Two questions I came out of the morning unsure about, and I’m listing them because “not sure” is more useful to you than a guess. One: what a skill is at the artifact level, and how you’d ship one you wrote yourself. Two: whose user the agent reads and writes as, and whether sharing rules and field-level security apply to what it can see and change.
That’s where I see the scaling problem, and it has nothing to do with request limits. Once the plugin is in, everyone starts building their own skills, and some of those will miss something or misread the org they’re pointed at. It’s inviting a Wild West campaign.
If you’re an admin reading this, the question to settle before the install is who gets to write a skill, who reviews it, and what it runs as. Whether it works was settled on stage.
Dave’s Day 1 dispatch, from the Architect’s Chair, is here.
Responses