Salesforce has shipped a ground-up rebuild of Slackbot — not an incremental upgrade, but a fundamentally different product running on Anthropic's Claude. The new version is generally available today to Business+ and Enterprise+ subscribers at no additional cost, with a full mobile rollout completing by March 3. The practical difference: instead of reminding you to add a teammate to a doc, it can pull customer records from Salesforce, cross-reference a usage dashboard you upload as an image, identify open enterprise deals worth targeting, draft a shared Canvas document summarizing the findings, and surface calendar availability for a stakeholder review — all in a single conversation thread.
The underlying architecture explains why this feels different from previous Slack AI features. Rather than a narrow rules engine, the new Slackbot connects an LLM to a robust search layer that spans Salesforce CRM records, Google Drive files, calendar data, and your organization's full Slack history. Permissions are enforced at the user level — the agent can only surface information a given employee already has access to, which is exactly why Beast Industries' security team signed off quickly during their pilot. That permission model is also why Salesforce doesn't train models on customer data: once something enters an LLM's weights, you can't selectively hide that knowledge from specific users.
Anthropics's Claude is the current engine, chosen partly because it was the only FedRAMP Moderate-compliant LLM option when Slack began building. That won't stay exclusive — Salesforce has confirmed Google Gemini integration is coming this year, with OpenAI remaining a possibility. The multi-model direction matters for builders: Salesforce is treating LLMs as interchangeable infrastructure (their CTO compared them to CPUs), which suggests the platform bet here is on Slack as the interface and data layer, not on any single model.
Internal numbers from Salesforce's 80,000-employee rollout are worth taking seriously: two-thirds of staff tried it, 80% of those became repeat users, and internal satisfaction hit 96%. Employees organically crowdsourced a shared document of 250+ useful prompts within five days of launch — a reliable signal of genuine utility rather than mandated adoption. External pilot customers including Beast Industries reported saving 90 minutes per day per user; Engine's SVP of Operations cited 30 minutes saved daily from reduced context-switching alone.
The competitive framing is straightforward: Microsoft Copilot lives inside Teams and Microsoft 365; Google Gemini is woven through Workspace. Salesforce's argument is that Slackbot has a context advantage — it's already embedded in years of your organization's actual work conversations and decisions, with no setup required from end users. The roadmap extends this further: Salesforce plans to make Slackbot an MCP (Model Context Protocol) client, positioning it as a coordination hub for third-party agents already appearing in Slack from Anthropic, OpenAI, Google, and Vercel.
One real cost consideration to flag: Slackbot itself is free on qualifying plans, but Salesforce is simultaneously tightening API access pricing for third-party data integrations. If your stack relies on tools like Fivetran to replicate Salesforce data to Snowflake, or on ChatGPT to query Salesforce records, you may face pressure to shift those workflows onto Salesforce's own Data Cloud and Agentforce products. Evaluate your data pipeline dependencies before assuming the total cost picture is unchanged.