If You Have Just One Minute

• Agent governance became non-negotiable. OpenAI’s rogue-agent incident widened to at least 10 more sites and triggered an EU incident report. Runtime controls, least-privilege access and a tested kill switch are now baseline requirements.
• Agents are gaining real-world action rights. Meta’s Muse sends emails, makes payments and books travel across connected apps — a preview of the permission, identity and auditability problems B2B will inherit.
• Enterprise AI is verticalizing. OpenAI launched ChatGPT for Financial Services with Morgan Stanley and Evercore, bundling trusted data, role-based access and audit logs. Generic copilots are losing ground to industry-specific systems.
• Economics are shifting from seats to outcomes. OpenAI’s enterprise revenue grew 32% month over month and hit parity with consumer, with specialized AI positioned around measurable results.
• Single agents are becoming fleets. Dynamics 365 Sales is making multi-agent Sales Qualification generally available, turning the core challenge into orchestration, ownership and conflict resolution.
• The conversational layer is absorbing CRM work. Sellers can now create accounts, contacts and opportunities directly from Copilot Chat.
• AI visibility is now a measurable channel. Adobe Brand Visibility tracks share of voice, source visibility and AI referral impact across 289M+ prompts, moving AEO/GEO into attribution territory.
• Enablement is stuck in the shallow end. Forrester finds fast AI adoption for content and automation, but lagging use in coaching, practice and competency development.
• Feature moats are shrinking. Salesforce, Intuit and ServiceNow fell 4–5% after the GPT-6 Astra release, a reminder that durable value sits in context, workflow and trust — not generic AI features.

The one-line takeaway: the winners this week were not the teams with the most agents, but the ones building the controls, context and measurement that let agents act safely and profitably.