Salesforce rolls out new Slackbot AI agent as it battles Microsoft and Google in workplace AI

Salesforce co-founder and Slack CTO Parker Harris likens the difference between the old and new Slackbot to a 'tricycle' versus a 'Porsche.' The old Slackbot performed simple algorithmic tasks like reminding to add a colleague or archive a channel; the new one is built on a large language model and a powerful search engine, connecting to Salesforce records, Google Drive files, calendar data, and years of Slack conversations. Harris emphasizes they are 'two different things' and the new version is 'brand new.'

Despite the complete architectural change, Salesforce kept the Slackbot brand because 'people know what Slackbot is.'

The new Slackbot currently runs on Anthropic's Claude model, primarily due to compliance requirements: Slack's business service needs FedRAMP Moderate certification, and Anthropic was the only vendor able to provide a compliant LLM at the time. However, Harris says they will support more providers this year, including Google Gemini and possibly OpenAI. He quotes CEO Marc Benioff's view that LLMs are becoming commoditized, calling them 'CPUs.'

Salesforce rolled out the new Slackbot internally to all 80,000 employees. Slack CMO Ryan Gavin calls it 'the fastest-adopted product in Salesforce history.' Internal data shows two-thirds of employees tried it, 80% continue to use it, and satisfaction is 96%, the highest of any Slack AI feature. Employees report saving 2 to 20 hours per week.

Adoption was largely organic. Within five days, employees created a Canvas called 'Most Stealable Slackbot Prompts,' which now has over 250 prompts. Principal UX Researcher Kate Crotty found that 73% of internal adoption was driven by social sharing, not top-down mandates.

In a product demo, Slackbot demonstrated cross-source information synthesis: analyzing customer feedback, uploading a dashboard image and correlating qualitative and quantitative data, then querying Salesforce to find suitable early-access customers, finally generating a Canvas document and scheduling a meeting. Slack CPO Rob Seaman notes that Canvas creation is an example of 'tool calling,' and more third-party tool calls will be added in the future.

Pilot customers include Beast Industries (MrBeast's parent company). CIO Luis Madrigal says deployment was 'one of the easiest,' with the security team quickly approving because Slackbot only accesses information the user already has permission to see. Beast Games marketing lead Sinan reports 'saving at least 90 minutes daily,' and creative lead Spencer calls it 'an assistant that watches things when I'm not there.' Other pilot customers include Slalom, reMarkable, Xero, Mercari, and Engine.

Harris explicitly states that Salesforce does not use customer data to train any models. He explains that if the model learned confidential conversations, there would be no control over who sees the answers. Slackbot only accesses information the user already has permission to see, a design that allowed security teams to approve quickly.