Note: The primary source for this article, VentureBeat, was unavailable at review time due to access restrictions. All claims below are attributed to VentureBeat’s reporting and have not been independently verified.
Salesforce has rebuilt Slackbot as an AI agent capable of searching enterprise data, drafting documents, and taking action on behalf of employees, according to VentureBeat. The new version is now generally available to Business+ and Enterprise+ customers and runs on Anthropic’s Claude, VentureBeat reports.
VentureBeat reports that Salesforce co-founder and Slack CTO Parker Harris compared the old and new Slackbot to a tricycle and a Porsche, respectively. The original Slackbot handled algorithmic tasks — reminders, channel archive suggestions, simple notifications — according to VentureBeat. The new version is built on a large language model and a search engine that VentureBeat reports can access Salesforce records, Google Drive files, calendar data, and Slack conversation history.
Architecture and model choice
VentureBeat reports that the decision to build on Claude was driven partly by regulatory requirements. Slack’s commercial service operates under FedRAMP Moderate certification to serve US federal government customers, and Harris is reported to have said Anthropic was the only provider that could offer a compliant model when Slack began building the new system.
VentureBeat reports Harris described plans to add other model providers within the year, citing Google’s Gemini as a strong candidate on performance and cost grounds, and OpenAI as a future possibility. On model commoditization, VentureBeat reports Harris described LLMs as analogous to CPUs. On training data, VentureBeat reports Harris said Salesforce does not train models on customer data, reasoning that there would be no reliable way to enforce differential access to a model trained on confidential conversations.
Internal rollout
VentureBeat reports Salesforce tested the new Slackbot internally across all 80,000 employees before the external launch. According to VentureBeat, Slack CMO Ryan Gavin said two-thirds of Salesforce employees tried it, 80% of those continued using it regularly, and internal satisfaction reached 96%. VentureBeat reports employees described saving between two and 20 hours per week.
VentureBeat also reports that about five days into the internal rollout, employees created a collaborative Canvas document called “The Most Stealable Slackbot Prompts,” which grew to over 250 prompts. VentureBeat reports Kate Crotty, described as a principal UX researcher at Salesforce, found that 73% of internal adoption was driven by social sharing. VentureBeat reports Gavin called it the fastest adopted product in Salesforce history.
What the agent does
VentureBeat describes a product demonstration by Slack’s product experience designer Amy Bauer, who showed Slackbot analyzing customer feedback: uploading an image of a usage dashboard, correlating data across sources, querying Salesforce to identify enterprise accounts with open deals, synthesizing results into a Canvas document, and finding calendar availability for stakeholders. VentureBeat reports Bauer described the system as comparing image content against simultaneously generated insights, not simply reading the image.
VentureBeat reports Slack CPO Rob Seaman described the Canvas creation as a signal of where the product is heading, with plans to add additional third-party tool calls over time.
VentureBeat also reports that Beast Industries, the parent company of YouTube creator MrBeast, was among external pilot customers. VentureBeat reports the company’s CIO, Luis Madrigal, described the rollout as among the easiest technology deployments he had managed in over two decades, with employees reporting savings of approximately 90 minutes per day.
VentureBeat reports Harris described the product’s broader role as the front door to the agentic enterprise, powered by Salesforce.