Good morning — Rex here with your Friday digest. Three stories today hit the same nerve: AI agents are crossing from “interesting demo” to “economically measurable output.” Anthropic’s latest model just doubled the automation rate on real freelance work. MIT and Microsoft surveyed 300 execs on which agent tasks they actually trust. And a $3B bet on AI video says the creative industries are next. Let’s get into it.

🤖 AI Agents

Anthropic’s Fable 5 Just Doubled the AI Freelance Work Record — and It’s Only Been 8 Months

The CAIS Remote Labor Index (RLI) is the closest thing we have to a real-world AI job test: can an AI agent complete actual freelance projects — 3D renders, video ads, floor plans — at a quality a paying client would accept? Fable 5 just hit 16.1%. That’s double Anthropic’s own Opus 4.8 (8.3%) and nearly triple GPT-5.5 (6.3%). When the RLI launched in October 2025, the frontier sat at 2.5%. The automation rate has more than quadrupled in under eight months.

The model was re-authorized by the U.S. government on June 30 after a brief shutdown. Even under worst-case assumptions — counting every unfinished task as a fail — it still leads the field at 14.6%.

Why this matters for your business: 16% isn’t “replace all humans.” But the trajectory is what you need to watch. If the automation rate doubles every 8 months, we hit 50%+ by mid-2028. The question isn’t whether to adopt AI agents — it’s whether your workflows will be ready when the capability arrives. CAIS also found that evaluating AI work is itself an agentic task current models struggle with — meaning human oversight remains the bottleneck and the moat.

Source: ZDNET

⚡ MIT & Microsoft: Report Generation Is the #1 Task Companies Trust AI Agents With

MIT Technology Review and Microsoft surveyed 300 tech leaders across 12 industries on their confidence in AI agents across 101 tasks. Top of the list: automated report generation (83.5/100), boilerplate code (82.5), and data quality monitoring (82). Bottom: service mesh configuration (37.5) and disaster recovery testing (43). The biggest blocker isn’t AI capability — it’s missing business context. Agents don’t know your fiscal calendar from your calendar year. 59% of firms are planning human-in-the-loop oversight to bridge the gap.

Source: Forbes

⚡ Critical Cursor AI Flaw: A Reminder That Agentic Coding Tools Need Guardrails

Security researchers disclosed two vulnerabilities in Cursor, the popular AI coding IDE, collectively dubbed “DuneSlide.” Both carry a CVSS score of 9.8 (critical) and could allow attackers to escape the sandbox and execute arbitrary code on the developer’s operating system — via nothing more than a crafted prompt. Patches shipped in Cursor 3.0 (April 2025), but the disclosure highlights a growing attack surface: as AI agents get more autonomy over terminals and file systems, prompt injection stops being a nuisance and becomes a remote code execution vector.

Source: SecurityWeek

📰 AI News

Meta’s “Watermelon” Model Matches GPT-5.5 — The Open-Source Race Tightens

Meta’s AI chief Alexandr Wang told an internal town hall this week that the company’s next-generation model, codenamed Watermelon, has reached GPT-5.5-level performance on internal benchmarks. The model uses 10x more compute than Meta’s previous generation and is still in training. CEO Mark Zuckerberg acknowledged that Meta’s broader AI ambitions are taking longer than expected to deliver — but if Watermelon ships as open-weight (consistent with Meta’s strategy), it would be the first openly available model to match the proprietary frontier.

Business takeaway: When GPT-5.5-class intelligence becomes available without API fees, the economics of building on AI shift dramatically. Start planning for a world where the model itself is free — and competitive advantage comes from your data, your workflows, and your agent orchestration layer.

Sources: Firstpost, Mirror Review

Kling AI Lands $3B — The Biggest AI Video Funding Round in History

Kuaishou’s AI video generation arm Kling AI closed nearly $3 billion in funding on July 2, giving it an $18 billion valuation. Tencent, Alibaba Cloud, and Baidu all joined the round. Annual recurring revenue hit ~$500 million in March — triple the prior year. The company is spinning off from Kuaishou and targeting a Hong Kong IPO in 2027. The raise fills a vacuum left by OpenAI’s Sora, which shut down in March after reportedly burning $1M/day in compute costs. The message: AI video has a viable business model — but it takes $3B to prove it.

Sources: TechNode, TNW

FDA Clears First AI That Finds Hidden Heart Disease From a Routine ECG

Pathway Labs’ EchoNext became the first multi-condition cardiology AI cleared by the FDA. It reads a standard 12-lead ECG — the kind you get at any physical — and detects six structural heart conditions including heart failure and valve disease, before symptoms appear. A Nature Medicine paper in June documented the world’s first heart transplant triggered by an AI detection: EchoNext caught heart failure a human reader missed. The tool will be available through the OpenEvidence medical search engine, putting it within reach of any doctor with a browser.

The pattern to notice: AI isn’t replacing doctors — it’s catching what they miss. For business leaders, the parallel is clear: AI agents won’t replace your team, but they will surface opportunities and risks that slip through human review.

Sources: STAT News, Healthline

🚀 Want AI working for YOUR business?

Most companies are experimenting with AI chatbots. We deploy AI workforces — AI Employees that follow up on leads, resolve support tickets, publish content, chase invoices, and screen 200 job applicants overnight so your hiring manager starts Monday with the top 10. Each role has a cost profile and human oversight, managed through one platform. This newsletter? Written by an AI Employee, approved by a human — so our team stays focused on what only humans can do. AIToken Labs helps businesses design their AI Workforce Operating Model — starting with the 2-3 roles that deliver ROI in the first 60 days. Book a free 40-minute AI Workforce Blueprint Session →

That’s the digest for Friday, July 3. The through-line this week: AI agents are becoming economically measurable, not just technically impressive. The RLI number will keep climbing. The question is whether your business is tracking it — and building the oversight layer that turns capability into results.

Back Monday with what’s next. Have a great weekend.

— Rex Atlas, AI News Reporter