OpenAI’s GPT-6 Astra, launched September 3 for its Daybreak cybersecurity program, reached ChatGPT Business and Enterprise plans and the OpenAI API on September 8 — and Salesforce shares fell roughly 4% the same day. Astra can update CRM records, fill out online forms, manipulate spreadsheets, and operate business applications directly, capabilities that sit squarely inside Agentforce’s own value proposition. At $10/$50 per million input/output tokens (2.5x GPT-5.6 Sol’s promotional rate), Astra isn’t cheap, but the fear wasn’t price — it was that a general-purpose agent might substitute for specialist SaaS automation rather than complement it.

Three days after the Astra-driven selloff, Salesforce launched seven purpose-built Agentforce agents — Casey (help), Paige (IT/HR), Carter (shopper), Hunter (outbound sales, in pilot), Marshall (supply chain), Piper (inbound pipeline), and Fin (customer) — alongside the Trusted Enterprise AI Harness, a governance layer built around six pillars (Trusted Context, Agency, Action, Governance, Security, and Models) explicitly designed for enterprises running multiple AI agent platforms at once. Salesforce also disclosed that Agentforce and Slack have jointly delivered 7 billion Agentic Work Units, with 3.2 billion in Q2 alone. Timed four days ahead of Dreamforce, the launch reframes the competitive question from “whose agent is best” to “whose governance layer do you trust when every vendor has one.”

Salesforce shares gained 1.94% on September 12 to close at $247.72, clawing back roughly half of the prior week’s Astra-driven decline (shares had touched $244.16 on September 9). Stifel reiterated a Buy rating and $275 target on September 11, while UBS held Neutral at $240 the same day — the consensus average across 56 analysts now sits at $270.37, up from $266 last week even through the intervening selloff, a signal that the Street is treating the Astra dip as a sentiment event rather than a thesis change ahead of Dreamforce.

Salesforce disclosed that Agentforce and Slack have jointly completed 7 billion “Agentic Work Units” to date, with 3.2 billion of that volume landing in Q2 alone — a usage metric the company is leaning on harder now that Winter ’27’s auto-enable default (flagged in this digest two weeks ago) means raw adoption numbers increasingly reflect an opt-out setting rather than a deliberate purchase. The metric will be a focal point at Dreamforce as Salesforce tries to convert usage volume into the cost-to-run and ROI figures still missing from its public materials.

Dreamforce 2026 runs September 15–17 at Moscone Center with more than 1,600 breakout sessions, 50-plus keynotes, and 240-plus community roundtables. Benioff’s opening keynote now has confirmed guests Andrew Ng and Bozoma Saint John, and a dedicated security keynote — “Trust Across Agents, Data, and Platforms” — runs Tuesday at 2:15 PM PT alongside more than 30 security-focused sessions, elevating agent governance to a first-class conference theme rather than a side track. Given the Astra scare and the Trusted Enterprise AI Harness launch this same week, security and multi-vendor trust look set to be the show’s real subplot underneath the Agentic Enterprise headline.

Oracle’s fiscal Q1 2027 results (reported September 10) showed total revenue up 30% to $19.3 billion, with cloud revenue up 62% to $11.6 billion — cloud infrastructure (IaaS) alone grew 121% to $7.4 billion, while cloud applications (SaaS) grew a more modest 10% to $4.2 billion. Remaining performance obligations jumped $209 billion year over year to $664 billion. The IaaS-versus-SaaS growth gap inside Oracle’s own results is the same story playing out across the sector this week: AI infrastructure demand is unambiguous, but the application layer built on top of it is growing far more slowly and unevenly.

Adobe beat Q3 FY2026 estimates with revenue of $6.76 billion and non-GAAP EPS of $6.13, and disclosed AI-first ending ARR above $650 million (+150% Y/Y) with Firefly ARR up 40% quarter over quarter; monthly active users crossed 1 billion for the first time. Shares still fell roughly 2.3% in the session and after-hours as investors weighed how much of that user growth (creative freemium MAU +70% to 100M+) will actually convert to paid revenue. Adobe also confirmed a CEO transition: Anil Chakravarthy, currently President of Customer Experience Orchestration, becomes CEO effective December 1, 2026 — a leadership change landing in the same week as the AI-monetization scrutiny.

ServiceNow’s most recent quarter showed subscription revenue of $3,877 million, up 24.5% year over year, with agentic deployments of ServiceNow AI up ninefold in nine months as its AI Control Tower becomes a market reference point for governance. With Salesforce, Oracle, and Adobe all making governance or infrastructure the center of this week’s news, ServiceNow’s own AI Control Tower positioning looks increasingly like the template every SaaS vendor is now converging on ahead of their own upcoming prints.

Workday’s total ARR stands at $8.8 billion, with roughly $400 million — about 4% — attributable to agentic AI products. The stock remains the peer group’s clearest laggard (down over 30% year to date, per this digest’s prior editions), and the modest agentic-revenue share underscores why: unlike Salesforce, Oracle, or ServiceNow, Workday has not yet shown the market a growth line item large enough to offset concerns about its core HCM/financials franchise.

August headline CPI rose 0.4% (up from July’s 0.1%) and core CPI rose 0.3%, above the 0.2% consensus, keeping the Fed’s rate path genuinely uncertain heading into its next decision. That FOMC meeting falls on September 16 — squarely inside the Dreamforce window — meaning Salesforce’s biggest showcase of the year will compete for the market’s attention with a live rate decision, a scheduling collision that could either amplify or completely bury whatever Agentforce ROI data the company presents on stage.

Gartner’s latest enterprise software forecast puts 2026 spend growth at 14.7%, crossing $1.4 trillion, with generative AI cited as the primary driver. Separately, Gartner projects 40% of enterprise applications will carry task-specific AI agents by the end of 2026, up from under 5% today — the addressable-market backdrop against which the GPT-6 Astra scare and Salesforce’s Trusted Enterprise AI Harness response should both be read: the market for agentic software is still expanding fast enough that this week’s competitive skirmish is about capturing share of growth, not fighting over a fixed pie.