Meta rebuilt its AI from scratch. Anthropic launched managed agents. Oracle fired 30,000 by email. MiniMax shipped a model that trains itself. Eight stories from the week AI infrastructure became a staffing decision.
Oracle fired 30,000 people by email at 6 a.m. to fund AI data centers. The same week, Anthropic launched a managed agent platform that eliminates months of infrastructure work, Meta shipped an entirely new model built from scratch under Alexandr Wang, and a Chinese startup released an AI that handles half its own training. The agentic era isn't arriving — it's restructuring headcount, rewriting org charts, and repricing the cost of intelligence itself. Eight stories from the week AI stopped being a product category and became a staffing decision.
Meta Rebuilt Its Entire AI Stack From Scratch.
Nine months after hiring Alexandr Wang and spending up to
on AI capex, Meta Superintelligence Labs shipped its first model — and it's not open-source. Muse Spark, code-named Avocado, is a natively multimodal reasoning model with tool-use and multi-agent orchestration built in. It ranked fourth on the Artificial Analysis Intelligence Index, accepts voice, text, and image inputs, and exposes two modes: Instant and Thinking, with a deeper Contemplating mode coming. The model will roll out across Facebook, Instagram, WhatsApp, Messenger, and Ray-Ban glasses in the coming weeks. Meta says an open-source version will follow — but the closed launch is the signal. Meta is competing head-to-head with OpenAI and Google on capability, not just distribution.
The closed launch is the tell. Meta spent years building open-source credibility with Llama — and the first model under Wang ships proprietary. That means they think Muse Spark is good enough to compete directly, and they want the distribution advantage before handing the weights to competitors. Watch whether the open-source version arrives intact or strategically crippled.
Anthropic Wants to Host Your Agents For You.
Anthropic shipped Claude Managed Agents on April 8 — a composable API suite that lets developers define tasks, tools, and guardrails while Anthropic handles the ugly parts: orchestration, sandboxing, error recovery, scaling, and monitoring. Pricing is
per session hour on top of standard token costs. Early adopters include Notion, Rakuten, and Asana, running production workflows in coding, task automation, and document processing. The strategic play is clear: Anthropic wants to own the infrastructure layer for enterprise agents the way AWS owns compute. If you're building agents, the question is no longer "which model" but "whose infrastructure."
This is Anthropic's AWS moment. Managed Agents turns Claude from a model into a platform — and platforms have lock-in that models don't. The $0.08/hour pricing is deliberately cheap enough to make "build it yourself" feel irrational. Every enterprise that deploys here becomes harder to migrate away.
Oracle Fired 30,000 People by Email at 6 A.M.
Oracle began its largest layoff in company history: up to
of its workforce — terminated via a form email sent at 6 a.m. simultaneously across the U.S., India, and other regions. The email, signed "Oracle Leadership," told workers their roles had been eliminated as part of a broader organizational change. Screenshots flooded social media within hours. The motive is pure arithmetic: TD Cowen estimates the cuts free up
in cash flow, money Oracle needs to fund its aggressive AI data center buildout. This came the same quarter Oracle posted
profit surge. The cruelty is the point — or at least the efficiency is.
"Following a review of the company's current business needs, a decision has been made to eliminate your role as part of a broader organizational change."
Last week we covered Oracle rebuilding Fusion around agents. This week we see how they're paying for it — with 30,000 jobs. The AI transition isn't free. The companies that handle it with transparency will keep their best people. The ones that send mass emails at dawn will learn that employer brand is a competitive asset they just torched.
OpenAI Loses Three Executives in One Week.
OpenAI's leadership bench took three hits in a single week. Fidji Simo, the company's product and business chief hired last May, announced medical leave for a relapsing neuroimmune condition. Greg Brockman will oversee product in her absence. COO Brad Lightcap was moved to "special projects" — reporting directly to Altman — with CRO Denise Dresser absorbing most of his responsibilities. And marketing chief Kate Rouch stepped down to focus on cancer recovery, with former Meta CMO Gary Briggs filling in on an interim basis. Three departures, two medical emergencies, one restructuring — and the company is months away from what would be the largest tech IPO in history.