What happened
Dreamforce 2026 runs September 15 to 17 at Moscone Center, with Benioff's opening keynote at 10 a.m. Pacific today. Salesforce confirmed a joint main-stage conversation with Anthropic CEO Dario Amodei to formalize how Claude, Agentforce, Slack, Data 360, and Headless 360 fit into a shared enterprise architecture. The company disclosed that Agentforce and Slack have collectively delivered 7 billion Agentic Work Units, with 3.2 billion booked in Q2. Salesforce in Claude is in select-pilot now and enters open beta this month. Analysts including TechRadar and MarketScale flagged the event as the first time a public software company has attached a hard, disclosable usage metric to AI-agent work — which forces every competitor to either match the number or explain why they will not.
Why Agentforce Agentic Work Units matters now
MagnetizeX builds founder visibility systems for B2B firms.
Software analysts have argued for two years about whether AI-agent revenue is real or bundled. A public disclosure of 3.2 billion units in a single quarter drags the conversation from opinion into accounting. If the average enterprise buyer wants to know whether the category is used or shelved, they now have a number that is quotable in a board deck. That reprices comparable AI-agent companies in the private market and pressures HubSpot, Microsoft, and Google to publish their own usage stats. It also gives GTM founders and CMOs a new procurement anchor: buyers can now ask any AI-agent vendor for a Work Units equivalent and treat non-disclosure as a red flag.
- Track how each AI-agent vendor responds to the Agentic Work Units disclosure by Friday
Watch which of Microsoft, HubSpot, ServiceNow, Google, and the private AI-agent challengers publish their own usage metric within 30 days. Non-response is a signal that usage is lower than expected and the category may consolidate faster than the funding round suggests. Founders who track this will get an early read on the M&A shortlist.
- Reprice comparable-company assumptions in your GTM planning
If Salesforce's Agentforce numbers hold, the AI-agent category is trading closer to usage-based software than to CRM software. Update your competitive slide and any board narrative that relied on private-market comps. The public-comp anchor just changed, and any pitch that references AI-agent TAM without citing this quarter's disclosure is now behind.
- Add a usage-metric ask to every AI-agent procurement conversation this quarter
When evaluating any AI-agent vendor, ask for their equivalent to Work Units — tasks completed, workflows executed, or human-hours displaced. Vendors that cannot answer are either not measuring or are hiding the number. Neither is a reason to buy, and the ask itself signals that your team has caught up to the market.
By the numbers: Q2 alone accounted for 3.2 billion of the 7 billion cumulative Agentic Work Units, implying Q2 delivered roughly 46 percent of all-time Salesforce agent work in a single 90-day window. Dreamforce 2026 hosts 1,600 breakout sessions, 50-plus keynotes, and 240 community roundtables over three days.
What to do this week
Watch the Benioff-Amodei main-stage session at 10 a.m. Pacific and note two things: which enterprise customers Salesforce puts on stage as Agentforce case studies, and whether the Work Units metric gets broken down by agent type. Both signal where the category revenue actually sits. Then, in your next investor update or board deck, replace any AI-agent TAM slide with the Agentforce Q2 usage number as the reference anchor.