Anthropic's annual conference last week shipped enterprise infrastructure rather than another headline model — Managed Agents, multi-agent orchestration, outcomes-as-rubric, a memory feature called dreaming, and a serious compute expansion. Most of the coverage reads like a product launch recap. Stephen reframes it as a P&L event and walks through the three-stage method for turning announcements like these into a workflow change a CFO will defend in the budget cycle.

What's covered

• What Anthropic actually shipped — Managed Agents, multi-agent orchestration, outcomes (rubric-based self-checks), the dreaming memory feature, and why the compute expansion is the silent variable that turns a fragile experiment into a budget line
• Why most enterprise AI rollouts stall — not a model problem, a sequencing ...