AI Agent Digest: Week 17, 2026 — Google Goes Full-Stack Agentic, OpenAI Ships Workspace Agents, Meta Cuts 10% for AI

Week 17 was the week the hyperscalers stopped pretending agents were a feature and started treating them as the entire product line. Google reframed its whole enterprise AI stack around agents. OpenAI launched persistent workers inside ChatGPT. Meta cut 8,000 jobs and pointed at AI. And somewhere in the middle, Cursor quietly raised at $50B. Here's the tour.

1. Google Cloud Next: the full-stack agentic bet

Google went all-in at Cloud Next 2026. The announcements: a no-code agent builder for Workspace, a rebranded "Gemini Enterprise Agent Platform" replacing Vertex AI, managed MCP servers across Google Cloud, a web-browsing agent called Project Mariner, an "Agent Inbox" where bots post status to humans, and production-grade Agent2Agent (A2A) support. Third-party models including Anthropic's Claude are now first-class inside Google's platform.

Hot take: This is the first platform move in the space that actually pattern-matches to AWS in 2008. Google stopped selling a model and started selling the infrastructure other people's agents run on. Anthropic is the tell — Google is hosting its fiercest model competitor because it understands the real margin is in the platform underneath, not the model on top. OpenAI is a SaaS company now. Google is a cloud.

2. OpenAI Workspace Agents go live in ChatGPT

OpenAI shipped Workspace Agents to Business, Enterprise, Education, and Teacher plans. These are persistent, multi-step workers with memory and schedules — running inside Slack, Salesforce, and Google Drive without a human pressing "go" between steps.

Hot take: OpenAI's product strategy is now transparent: vertically integrate the agent experience inside ChatGPT, keep developers upstairs in the API tier, and lock enterprises into the workspace layer before Google's open platform eats them. Two years ago we were arguing about whether agents were real. Now OpenAI is shipping them as a Slack bot and calling it "digital coworkers." The branding war is over. Execution is the fight.

3. Meta cuts 10% (8,000 people) and blames AI. Then leaks a surveillance tool.

Meta announced layoffs of 10% of its workforce — roughly 8,000 employees — explicitly framed as an AI efficiency push. In the same week, internal memos revealed a new program called the "Model Capability Initiative" that captures keystrokes and mouse clicks from Meta employees' work computers to train AI agents.

Hot take: The pattern is clearer every quarter: the companies cutting the hardest are the ones also building the agents that replace the cut roles, trained on the data of the people doing those roles until the day they're laid off. Call it what it is — the most expensive employee offboarding program in history, and the offboardees are training their own replacements on the way out. If your job touches a keyboard at Meta, assume you're a data source.

4. Merck + Google Cloud announce a $1B agentic partnership

Merck and Google Cloud committed to a partnership valued at up to $1B to deploy an agentic platform across R&D, manufacturing, commercial, and corporate functions. This is pharma's first nine-figure agentic bet.

Hot take: Pharma is traditionally the industry most terrified of putting AI near anything regulated. Merck writing a $1B check to Google Cloud means the compliance story is no longer blocking — it means Google convinced Merck's general counsel that agentic workflows can be audited, gated, and logged well enough for FDA territory. If it works at Merck, every regulated vertical watches next.

5. Deloitte spins up a dedicated Google Cloud Agentic Transformation Practice

Deloitte announced an end-to-end agentic transformation practice built on Google's stack — designed to help Fortune 1000 companies deploy agents "securely and at scale."

Hot take: When a Big Four creates a named practice with Google branding, the implementation margin is locked in for the next three years. The Fortune 1000 will not buy agents; they will buy Deloitte-led agent rollouts, because their risk committees cannot approve anything else. This is the ERP rollout playbook with a new name. If you're building for enterprise, you're selling to Deloitte before you're selling to the customer.

6. Cursor raising $2B at a $50B+ valuation

Cursor is in talks for a $2B round at a $50B+ valuation, co-led by a16z with Nvidia and Thrive Capital. The AI-coding-agent market just minted its clear #2 behind OpenAI in dollar terms.

Hot take: $50B for a coding IDE is insane until you realize Cursor is not an IDE — it's a durable position on where developer attention lives. The question isn't whether the valuation is justified today; it's whether "the place engineers write code" is more valuable than the model doing the writing. The bet is that attention beats capability, and so far the bet is holding.

7. OpenAI's seventh acquisition of 2026

Hiro is winding down, with its ~10-person team joining OpenAI on April 20. That's OpenAI's seventh acquihire of 2026, already nearly matching its entire 2025 total of eight with eight months left in the year.

Hot take: OpenAI is buying specific humans at a tempo that only makes sense if they're losing a hiring war internally. Each of these acquisitions is a team that almost certainly turned down a direct offer. The story of 2026 is not "AI replaces workers" — it's the most violent talent war in tech history for the three thousand people on Earth who can actually ship an agent. If you're one of them, you have never had more leverage.

8. The real adoption numbers are out — and they're big

Two datapoints worth their own section. GE Appliances reported 800+ Google AI agents running across manufacturing, logistics, and supply chain. KPMG hit 90% Gemini Enterprise adoption among employees with 100+ agents deployed in the first month.

Hot take: The story everyone's been writing for two years — "enterprises are still piloting" — just ended. 90% adoption in month one at a Big Four firm isn't a pilot; it's production. The question has shifted from "will enterprises deploy agents" to "how many, how fast, and with what governance." If your SaaS pitch still involves the word "pilot," your deck is already out of date.

What we're watching next week

• Microsoft Build 2026 is next month but pre-announcements start now — expect a response to Google Cloud Next around Azure AI Foundry and Copilot agent orchestration.
• Anthropic Managed Agents pricing is rumored to ship; watch for whether they undercut OpenAI Workspace Agents on per-seat or shift to a compute-based model.
• EU AI Act enforcement calendar — the next high-risk classification review window is late April; agentic systems in financial services and HR are expected to be on the list.

Bottom line

Week 17 was the week "agents" graduated from the press-release tier to the P&L tier. Google, OpenAI, and Meta made moves that will show up in 10-Ks. Merck, Deloitte, and KPMG made moves that will show up in your procurement cycle. The infrastructure layer is consolidating around two platforms (Google's full-stack and OpenAI's vertical), the integrator layer is consolidating around Deloitte, and the workforce implications are now officially in the numbers rather than in the forecasts.

If your 2026 plan still reads like "we should try some AI pilots," rewrite it this weekend.

Want to test the most advanced AI employees? Try it here: https://Geta.Team