Salesforce didn’t just update Slack last week. It declared that the chat window — not the CRM, not the ticketing system, not the database — is where enterprise AI gets controlled.
The April 1 announcement shipped 30 new AI features, but the one that matters most barely made the headlines: Slackbot is now a Model Context Protocol client. That means it can connect to and coordinate external services, agents, and tools — including but not limited to Salesforce’s own Agentforce. As Rob Seaman, Slack’s interim CEO, put it: the bot finds “the most relevant and efficient path for the information, without human intervention.”
This matters because it tests one of the unresolved questions in the software commoditization thesis: when AI does the work, who owns the customer?
The orchestration bet
The traditional SaaS stack assumed each tool owned its own interface. You logged into Salesforce to manage deals, Freshdesk to handle tickets, Workday to approve PTO. The tool was the interface, and the interface was the moat.
Salesforce is betting that’s over. If Slack can dispatch work to any agent in the enterprise — route a support case to an AI resolution engine, update a CRM record from a conversation, trigger a finance workflow from a meeting summary — then the tools underneath become interchangeable backends. The value migrates from the tool to the dispatch layer.
Marc Benioff framed this in distribution terms: a million businesses already run on Slack, with 2.5x revenue growth since the acquisition. That’s the installed base you need to credibly claim the orchestration role. And the MCP integration isn’t just connecting to Agentforce — it’s designed to route to “any agent or app in your enterprise,” which positions Slack as a protocol-level interface, not just a Salesforce feature.
What this means for the tools underneath
The timing of this announcement sits uncomfortably next to what’s happening at the other end of the SaaS stack.
On March 19, Oppenheimer downgraded Freshworks from Outperform to Market Perform, removing its $15 price target. The core concern: Freshworks has weak “AI defensibility” and its seat-based pricing model faces “brittle growth and weakening pricing power.” Oppenheimer’s explicit advice: favor companies moving toward “tokenization or consumption models” over those clinging to per-seat.
Freshworks has tried to adapt. Its Freddy AI charges $0.10 per session — but per session, not per resolution. Compare that to Intercom’s Fin at $0.99 per resolved conversation, where better AI performance directly translates to less cost per outcome. The pricing structure reveals the strategic position: Intercom sells the outcome, Freshworks sells the attempt.
Now layer on the Slack MCP announcement. If the dispatch layer can route a customer support request to whichever AI agent handles it best — Intercom’s Fin, a custom Agentforce agent, or an open-source resolution engine — then the individual tool’s brand matters less than its resolution rate and cost per outcome. The orchestration layer sets the terms. The tools compete on performance.
The moat question
This doesn’t mean every SaaS tool gets commoditized overnight. The counter-evidence is real: Freshworks isn’t just a UI layer. It holds customer history, SLA configurations, compliance rules, routing logic built up over years of enterprise deployments. That data context doesn’t evaporate because Slack learned MCP.
And Salesforce’s own history suggests orchestration layers aren’t automatic wins. The company paid $27.7 billion for Slack in 2021. Five years in, it’s still working to justify that price — Benioff’s “2.5x revenue growth” claim sounds impressive until you remember Slack was doing about $1.3 billion in annual revenue at acquisition, which would put it around $3.3 billion now against a $27.7 billion purchase price.
The real question is whether MCP — as an open protocol, not a Salesforce proprietary standard — enables a dispatch layer that any platform could implement. If Microsoft builds the same MCP routing into Teams, if Google does it in Workspace, then the orchestration layer itself commoditizes. Salesforce would have bet the interface on an open standard that any competitor can match.
What this tells us about the hypothesis
The dispatch layer thesis adds a new wrinkle to software commoditization. The original hypothesis focuses on AI creating software faster, collapsing margins. What Salesforce is testing is subtler: AI doesn’t just replace the software — it replaces the interface to the software. When users stop opening individual tools and start routing everything through a single AI-mediated surface, the tools lose their relationship with the customer.
This is verification with a caveat. The hypothesis holds stronger for point-solution SaaS (ticketing, scheduling, basic CRM) that lack proprietary data gravity. It holds weaker for systems of record with deep compliance and integration moats. And it holds differently for the orchestration layer itself — which may capture value precisely because it commoditizes everything below it.
Watch Freshworks’ next earnings for seat count trends vs. AI session volume. Watch whether Microsoft ships MCP integration for Teams (rumored for Build 2026). And watch whether the “dispatch layer” produces actual enterprise adoption numbers, or remains a keynote demo.
The software stack is getting a new layer, and it sits on top of everything.
Sources: Salesforce/Slack announcement via MarTech, TechCrunch, American Bazaar; Freshworks downgrade via MarketBeat.