AI Agent Digest: Week 9, 2026, Cowork, Cloud Agents, and Custom Agents: Everyone Shipped the Same Week

Three separate companies shipped autonomous agent products on the same day. That's not a coincidence,it's a land grab. Here's everything that happened in AI agents this week.

1. Anthropic Launches Claude Cowork,The "Office Worker" Play

Anthropic unveiled Claude Cowork on February 24, connecting Claude directly to Google Drive, Gmail, DocuSign, and FactSet. Available now to Pro plan users on macOS.

Hot take: While everyone else is building AI for developers, Anthropic just pointed Claude at the 90% of knowledge workers who've never opened a terminal. Cowork isn't competing with Cursor,it's competing with your company's next hire. The timing, right after Opus 4.6's 1M-token context window and Sonnet 4.6's agent planning upgrades, makes this feel less like a product launch and more like a calculated three-punch combo.

2. OpenAI Forms "Frontier Alliances" with McKinsey, BCG, Accenture, and Capgemini

OpenAI announced multiyear partnerships with four of the world's largest consulting firms to deploy its Frontier enterprise agent platform. BCG and McKinsey handle strategy; Accenture and Capgemini handle implementation.

Hot take: OpenAI basically admitted that technology alone doesn't sell to enterprises,relationships and implementation muscle do. Paying McKinsey to explain your product to Fortune 500 CIOs is the most expensive go-to-market strategy in tech history, and also probably the most effective. The real story? Frontier works with agents from Google, Microsoft, and Anthropic too. OpenAI is positioning itself as the enterprise agent operating system, not just an LLM provider.

3. Cursor Ships Cloud Agents,35% of PRs Now Written by Bots

Cursor launched Cloud Agents that run on isolated Linux VMs, onboard to your codebase, write code, test it, and deliver merge-ready pull requests. Users can trigger agents from the web, desktop, mobile, Slack, or GitHub. The $29.3B startup says 35% of its pull requests are now fully agent-generated.

Hot take: The "35% of PRs" stat is the number that should keep every dev tools company awake at night. Cursor isn't selling AI-assisted coding anymore,it's selling autonomous software delivery. When a third of your code changes come from machines running on cloud VMs you've never logged into, the distinction between "AI-assisted developer" and "developer-supervised AI" starts to blur fast.

4. Notion 3.3: Custom Agents That Work While You Sleep

Notion released Custom Agents on February 24,autonomous AI teammates that run 24/7, triggered by schedules or events. Early testers created 21,000 Custom Agents, and Notion itself runs 2,800 internally. They work across Notion, Slack, Mail, and Calendar. Free to try until May 3.

Hot take: Notion just turned every project manager into an agent builder. No code. No API keys. Just describe the job and set a schedule. 21,000 custom agents in early testing tells you the demand was always there,people have been waiting for "set it and forget it" automation that actually works. The pricing model (credits at $10 per 1,000) will be the real test. If agents run too hot, you'll get surprise bills. If they're too conservative, they won't be useful.

5. Airbnb Quietly Deploys AI Support Agent,Resolves a Third of All Issues

Airbnb rolled out an AI assistant in customer support across the U.S., Canada, and Mexico. The system already resolves approximately one-third of all customer issues before a human agent gets involved.

Hot take: "Quietly deployed" is doing a lot of heavy lifting here. Airbnb didn't announce this with a press release or a keynote,they just turned it on and measured the results. One-third resolution rate without human intervention is the benchmark every customer service team will now be measured against. The silence is the strategy: if customers don't notice, it's working.

6. Burger King Deploys OpenAI-Powered "Patty" Headsets at 500 Locations

Burger King is rolling out OpenAI-powered headsets called "Patty" across 500 locations. The system flags low inventory, removes sold-out items from digital menus in real-time, and alerts managers to maintenance issues.

Hot take: Forget the enterprise deals and developer tools for a second. AI agents just entered fast food. Patty isn't a chatbot,it's an operations manager that runs on a headset. When your local Burger King has more AI integration than most Series B startups, you know the technology has crossed from "cutting edge" to "utility." The name "Patty" is also chef's kiss for branding.

7. Jump AI Raises $80M Series B,1 in 10 U.S. Financial Advisors Now Use It

Jump closed an $80M Series B led by Insight Partners on February 19, bringing total funding to $105M. The AI platform for financial advisors has scaled from zero to 27,000 advisors in under two years, adding 2,000+ per month.

Hot take: Nearly one in ten U.S. financial advisors using the same AI platform in under two years is the kind of vertical adoption that horizontal AI companies dream about. Jump proves the playbook: pick a profession, understand their workflows deeply, build agents that fit those workflows exactly. Not "AI for everyone",AI for financial advisors, specifically. The $80M also signals that vertical AI agent companies are commanding real venture dollars, not just horizontal platform plays.

8. UC Berkeley Drops 67-Page Agentic AI Risk Framework

UC Berkeley's Center for Long-Term Cybersecurity published the Agentic AI Risk-Management Standards Profile, a 67-page framework extending NIST's AI Risk Management Framework specifically for autonomous agents. The framework defines six autonomy levels from L0 (full human control) to L5 (full autonomy, human as observer).

Hot take: The L0-to-L5 autonomy scale is the most useful thing to come out of AI governance this year. Instead of arguing about whether agents are "safe" or "dangerous" in the abstract, we now have a shared vocabulary: "This agent operates at L2" means something specific. Every enterprise deploying agents should be classifying them on this scale yesterday. The framework also flags "self-proliferation capabilities" as a risk category, which sounds like sci-fi until you realize most agent frameworks already let agents spawn sub-agents.

What We're Watching Next Week

• MWC 2026 kicks off,Qualcomm released a Networked Agentic AI Index ahead of the event. Expect telecom-specific agent announcements.
• Claude Cowork adoption numbers,How fast do non-technical users adopt enterprise agents?
• Cursor vs. Anthropic vs. OpenAI,The coding agent battle is now a three-front war.
• Notion's agent pricing reality check,Free trial ends May 3, but early usage patterns will tell the story.

Bottom Line

February 24, 2026 will be remembered as the day everyone shipped agents simultaneously. Anthropic, Cursor, and Notion all launched autonomous agent products on the same date. OpenAI brought in the consulting big guns the day before. Meanwhile, Airbnb and Burger King proved that agents aren't just for tech companies anymore.

The pattern is clear: AI agents are no longer a product category,they're a feature of every product. The question has shifted from "should we build agents?" to "how fast can we ship them?"

If your business still thinks of AI as "that chatbot on the website," this was the week that should change your mind.

Want AI employees that actually execute tasks,not just answer questions? Geta.Team deploys autonomous AI workers with persistent memory, real email addresses, and custom skills that adapt to your business. Start in 5 minutes.