Salesforce Just Gave Slackbot a Brain Transplant, and It’s Actually Good

Salesforce finally did what Slack should have done years ago: they killed the old Slackbot and built a new one that actually does something useful.

The original Slackbot was basically a glorified to-do list. Remind someone to add a doc. Suggest archiving a channel. That’s it. It was fine for 2014, but in 2026, that’s embarrassing.

The new version, which just hit general availability for Business+ and Enterprise+ customers, is a completely different animal. It runs on Anthropic’s Claude, can search across Salesforce records, Google Drive, calendars, and years of Slack conversations, and actually takes actions on your behalf.

Parker Harris, Slack’s CTO and Salesforce co-founder, described the upgrade in classic Harris bluntness: “The old Slackbot was, you know, a little tricycle, and the new Slackbot is like, you know, a Porsche.”

I’d argue it’s more like swapping a tricycle for a self-driving car, but I’ll let the metaphor slide.

Why Claude, and why that matters

Slackbot runs on Anthropic’s Claude, and Harris was upfront about why: compliance. Slack’s commercial service carries FedRAMP Moderate certification for U.S. government customers, and when they started building, Anthropic was “the only provider that could give us a compliant LLM.”

But that exclusivity is temporary. Harris confirmed they’ll support additional providers this year, specifically mentioning Google’s Gemini (“performance is great, cost is great”) and leaving the door open for OpenAI.

This aligns with what Marc Benioff has been saying for a while: LLMs are becoming commodities. Harris called them “CPUs” — interchangeable components you plug in based on the job. I think that’s optimistic, but it’s the right strategy for a platform play like Slack.

More importantly, Harris was crystal clear on training data: Salesforce does not train any models on customer data. “If we trained it on some confidential conversation that you and I have, I don’t want Carolyn to know — if I train it into the LLM, there is no way for me to say you get to see the answer, but Carolyn doesn’t.”

That’s the right answer, and it’s refreshing to hear a vendor say it without hedging.

The internal numbers are hard to ignore

Salesforce rolled this out to all 80,000 employees months ago. Ryan Gavin, Slack’s CMO, called it “the fastest adopted product in Salesforce history.”

The stats: two-thirds of employees have tried it. 80% of those users keep using it regularly. Internal satisfaction hit 96% — the highest for any AI feature Slack has shipped. Employees report saving between two and twenty hours per week.

Twenty hours a week. That’s not a typo. If that holds up in real-world use, this thing pays for itself in days.

What’s more interesting is how adoption spread. Kate Crotty, a principal UX researcher, found that 73% of internal adoption came from social sharing, not management mandates. Employees created a “Most Stealable Slackbot Prompts” Canvas internally, which now has over 250 prompts.

That’s the kind of organic adoption you can’t buy. It suggests the tool actually solves real problems instead of being another AI wrapper nobody asked for.

What it actually does

During a demo, Amy Bauer showed how Slackbot can synthesize information across multiple sources. She asked it to analyze customer feedback from a pilot program, uploaded a usage dashboard image, and had Slackbot correlate the two.

That’s the kind of cross-system intelligence that enterprise software has promised for decades but rarely delivers. If Slackbot can consistently pull data from Salesforce, Google Drive, Slack messages, and calendars into coherent answers, it might actually justify the “agentic AI” hype Salesforce is pushing.

Salesforce is positioning this as “the front door to the agentic enterprise,” which is marketing speak, but there’s real substance underneath. The old Slackbot couldn’t even search your own messages. This one can search everything your company has stored across multiple platforms.

The competition isn’t standing still

Microsoft has Copilot in Teams. Google has Duet AI in Workspace. Both are pushing hard on workplace AI agents. Salesforce needed this win, and the internal numbers suggest they might have it.

But the real test is external adoption. Enterprise customers are skeptical of AI features after years of overpromising and underdelivering. If Slackbot actually saves people 2-20 hours per week, it’ll sell itself. If it’s another half-baked assistant that hallucinates meeting notes, it’ll be dead on arrival.

I’m cautiously optimistic. The architecture is solid, the compliance story is clear, and the internal adoption is genuine. But Salesforce has a long history of shipping impressive demos that get bogged down in enterprise deployment.

For now, though, this is the most interesting thing Slack has done since Salesforce bought it.