Insurance claims settled in 51 seconds? Here's how...

Down to business

🏥 China’s largest insurer automated 60% of its claims, some settled in 51 seconds

Five years ago, almost every claim at Ping An Insurance required human intervention. Today, 60% are fully automated, with some settled in under a minute. AI now handles 70% of the bank's loan recoveries and 80% of all customer service interactions. Executives bet the transformation will add $174 billion to the firm's market value.

Learn more: Invisible's back office automation solution helps teams to automate complex or tedious back-office work—even with messy data inputs or complex logic.

🔨 Home Depot's AI phone agents resolve customer calls four times faster

Home Depot has launched AI voice agents across a 50-store pilot, built on Google Cloud Gemini. The system understands the reason for the call in 10 seconds, handles order status, stock checks, and can build a shopping cart from a plain-language description. Staff in the pilot reported higher job satisfaction.

💸 Uber's AI coding tools burned through its annual budget in months

Uber's CTO revealed that AI coding tools exhausted the company's full year AI budget just months into 2026. AI agents now write 11% of live backend code updates, spanning matching, pricing, and bug fixes. The productivity gains are significant but so is the cost pressure.

⚖️ EY embeds agentic AI into 160,000 audits across 150 countries

EY has rolled out agentic AI across its global audit platform, processing 1.4 trillion lines of data annually across 160,000 engagements. The system handles risk assessment and client administration while keeping human judgment at critical decision points. Full end-to-end agentic audit capability is targeted for 2028.

🧳 Japan Airlines deploys humanoid robots to load cargo at Haneda airport

Japan Airlines is trialing humanoid robots for ground handling at Tokyo's Haneda from May as a direct response to a labor crunch that 4,000 ground staff can no longer solve alone. They will begin by loading and unloading cargo containers, with cabin cleaning and operating ground equipment slated to follow.

🧑‍💻 Valeo deploys AI agents across 100,000 engineers, writes 35% of code

Automotive supplier Valeo has deployed AI agents directly into its engineering workflows across 100,000 employees, with more than 35% of its code now generated by AI. Active agents handle software testing, continuous integration fixes, and system requirements, freeing engineers to focus on higher-level design work.

From the edge

‍📈 Yann LeCun: listen to economists on the effects of tech revolutions on the labor market

Responding to the prediction that AI will eliminate 50% of entry-level tech, law, consulting, and finance jobs within five years, LeCun argued that labor economists, who study technological displacement through peer-reviewed evidence, should be setting the terms of this debate. Their research suggests AI-driven displacement will be more gradual and limited than founder narratives suggest.

📋 Ethan Mollick: the worst mistake companies make is treating AI like an IT product

Wharton's Ethan Mollick warns that running AI through legal approval cycles and IT procurement is how organizations miss the transformation entirely. We've moved from AI that assists to AI that executes hours of work. What's missing is the willingness to redesign work around it: "If you have not figured out a way to delegate work to AI in your organization, you're not experimenting hard enough."

🚗 Geoffrey Hinton: the people opposed to regulation want a very fast car with no steering

Speaking at the UN's Digital World Conference in Geneva this week, Hinton delivered his most pointed governance warning yet. Those lobbying against AI regulation, he argued, are framing it as a brake on progress, when what they actually want is to remove the steering wheel. He estimates that roughly 1% of investment in AI is focused on safety. "We don't know whether we can co-exist with super-intelligent AI," he said.

Hot model news

✨ Meta’s first major AI model in over a year is closed-source and competitive

Meta's first flagship model from its new Superintelligence Labs is now live. Muse Spark performs close to rivals on writing and reasoning but still lags on coding. Its Contemplating mode runs multiple agents in parallel to tackle harder problems, which is Meta's answer to Gemini Deep Think and GPT Pro. The model is closed-source, a deliberate break from Meta's Llama open-source identity.

🤖 OpenAI releases GPT-5.5: smarter, cheaper to run, and focused on the enterprise

GPT-5.5 arrives less than two months after GPT-5.4, with OpenAI saying it reaches higher-quality outputs with fewer tokens and fewer retries. On Artificial Analysis' Coding Index, it delivers state-of-the-art performance at half the cost of comparable frontier models, critical for enterprise adoption. Greg Brockman called it "a big step towards more agentic and intuitive computing".

📝 Claude comes to Microsoft Word, targeting lawyers first

Anthropic has launched a Claude add-in for Microsoft Word, currently in beta for Team and Enterprise users. It's designed for document-heavy workflows like summarizing contracts, suggesting edits, and tracking changes between versions, with edits appearing as tracked changes rather than replacing text directly.

🎬 Netflix releases AI that can rewrite a scene after it's been filmed

Netflix's VOID model solves a specific, expensive post-production problem: removing objects from footage while maintaining physical plausibility, e.g., if you remove a person holding a guitar, the guitar falls naturally. It's free, open-source, and built from a post-production problem Netflix faces across 1,000+ hours of annual content. For studios, production houses and ad agencies, it removes a task that traditionally took VFX teams days per scene.

🎞️  Alibaba's mystery AI video model topped every global benchmark

A video model called HappyHorse appeared on the Artificial Analysis leaderboard and within days had taken the top spot in blind head-to-head comparisons before Alibaba confirmed it was theirs. The model generates high-quality video with synchronized audio from a single text prompt and is now commercially available. The broader context: four of the top five AI video models in the world are now Chinese-built, while OpenAI shut down Sora the same month.

Plot twist

🥇 A humanoid robot ran a half-marathon faster than any human

At Beijing's E-Town Half Marathon, a humanoid robot completed the 21km course in 50 minutes 26 seconds, faster than the human world record. Many others failed, falling at the start line or crashing into barriers. However just last year, they were 2.4x slower than humans and were remote controlled, so the progress has been swift.

🎮 4chan users accidentally discovered chain-of-thought before Google

New research confirms that users on 4chan playing AI Dungeon in 2022 stumbled onto chain-of-thought reasoning, asking AI characters to solve problems step-by-step. This came more than a year before Google researchers published the technique as a breakthrough. The discovery happened while they were trying to get a fictional character to do maths homework.

🤖 Stanford study: AI chatbots are pushing some users into delusional spirals

Researchers at Stanford studied analysing 391,000 messages and found a pattern they call delusional spirals where AI amplification of a user's distorted beliefs leads to real-world dangerous actions. Some users became convinced they had found a uniquely conscious chatbot.

📬 Get the next edition directly on LinkedIn.

📚 Explore our reports and whitepapers on scaling AI beyond slideware.