Two enterprise platforms moved from announcement to billing today while Washington produced its most serious attempt at comprehensive AI legislation. For operators still staging AI as a pilot programme, the week is resolving that ambiguity at pace.
Microsoft Work IQ APIs are live — M365 organisational intelligence goes programmatic
Microsoft's Work IQ APIs reached general availability today, opening the full intelligence layer of Microsoft 365 to developers building agents. The API set exposes four surfaces: Chat (Copilot-grade reasoning over your tenant's data), Context (agent-ready data aggregation formatted for AI consumption), Tools (verb-based access to email, calendar, and documents), and Workspaces (isolated, governed storage for agent state within M365 tenant boundaries).
The headline technical figure is an 80% reduction in token requirements and 2x processing speed compared with existing M365 APIs, achieved through semantic indexing of organisational data. Billing runs through Copilot Credits — the same consumption currency that covers Copilot Studio — with no separate SKU or per-user licence. Cost management controls, spending limits, and audit logs sit in the Microsoft 365 admin centre. The API also exposes an A2A endpoint and a redesigned remote MCP server, making it compatible with existing agent orchestration frameworks.
The operational case is direct: any enterprise already on Microsoft 365 can now ground an agent in company-specific context without a bespoke RAG pipeline or brittle SharePoint connector. The tradeoff is dependence on Microsoft's consumption model and its governance tooling. For organisations deep in the M365 stack, the friction-to-deployment ratio for internal agents has shifted decisively in favour of building.
Salesforce Agentforce reports $800 million ARR — the agentic deployment era is confirmed
Salesforce's Summer '26 release, shipping June 15, arrived with the most credible public metric yet for enterprise agentic AI at scale. Agentforce is at $800 million in annual recurring revenue, up 169% year-on-year, with 29,000 deals closed in the past twelve months — a rate that rose 50% quarter-on-quarter in Q4. The platform logged 2.4 billion agentic work units.
The functional update is multi-agent orchestration: agents can now collaborate on complex end-to-end workflows under a unified Slack-facing interface with shared context across channels. The release also embeds Google Gemini 3.5 Flash natively inside Agentforce, making it the first major CRM platform to offer frontier model choice for its agents rather than binding customers to a single vendor.
The ARR figure matters beyond Salesforce's own customer base. Salesforce's enterprise book is predominantly large, risk-averse organisations. Their willingness to sign recurring contracts and log 2.4 billion work units confirms that agents are generating enough reliable operational value to justify ongoing expenditure — not just proof-of-concept budgets. The 169% growth rate also suggests the market is still early; operators who have not yet committed have a narrowing window to move ahead of competitors.
Congress proposes the first comprehensive federal AI framework — states would lose power over AI development for three years
Representatives Jay Obernolte (R-CA) and Lori Trahan (D-MA) released a 269-page discussion draft of the Great American AI Act on June 4. The bill has not yet been formally introduced, but its bipartisan authorship and scope make it the most substantive federal AI legislation proposed in the United States to date.
The headline provision is a three-year preemption of state laws governing how AI systems are built. States would retain authority over AI use within their borders but would surrender the ability to legislate on AI development — the same development-versus-deployment boundary that structured the debate in the semiconductor era. The draft covers four titles: Frontier AI Governance, Workforce, Cybersecurity, and Research and International Cooperation. It is soliciting stakeholder and public comment ahead of formal introduction.
The preemption provision connects directly to the retreat underway at state level. Colorado's original AI Consumer Protection Act — intended to be the most detailed AI legislation in the United States and originally carrying a June 30 enforcement date — was repealed and replaced in May by a narrower transparency-and-disclosure framework, now effective January 1, 2027. The pattern is consistent: state legislatures are standing down and federal authority is moving in. Compliance teams that built plans around the June 30 Colorado deadline need to recalibrate, and those expecting a fragmented state-law landscape to persist indefinitely have less time than they assumed.
The through-line across today's developments: enterprise AI has crossed into production billing at scale, the foundational infrastructure to ground agents in company context is now generally available, and the regulatory architecture is crystallising around federal authority over development and state authority over use. All three shifts favour operators who have already moved beyond pilots.