Your cursor has done one thing for 50 years. Google just decided it should do a lot more.

Yesterday in AI | Thursday, May 14, 2026

Your cursor has done one thing for 50 years. Google just decided it should do a lot more.

Google just rebuilt the cursor. Amazon handed your shopping cart to an agent. And workers at some of the world's biggest tech companies are already gaming the AI metrics their bosses are tracking.

Today's episode covers seven stories: from a $2.1 billion bet on AI drug discovery, to a lawsuit accusing Meta of treating scam ads like a revenue instrument, to Hollywood's biggest names building something more concrete than a petition. There's also a product launch for legal teams that could finally make the "AI in our workflows" conversation real. Lots to get into.

Remember to subscribe, rate, and share this podcast if you like it!

Transcript

SPEAKER_00 0:05

Hi folks, this is Yesterday in AI, your daily digest of everything happening in the world of AI in 10 minutes or less. I'm Mike Robinson. It's Thursday, May 14th, and Google spent Tuesday reinventing the cursor. Amazon handed your shopping cart to an AI agent, and it turns out workers at some of the biggest tech companies are already gaming the AI metrics their bosses are tracking. Let's get into it. We'll start in Mountain View, where Google kicked off its one-week countdown to Google IO, its annual product showcase, with the biggest bet it's made in about 15 years. They're calling it Google Book, a new laptop category built from scratch around Gemini AI, launching this fall in partnership with Acer, Asus, Dell, HP, and Lenovo. It's a direct challenge to the MacBook and Google knows it. The machines run a merged version of Android and Chrome OS codenamed Aluminum, so you get Android apps, full phone sync, and Chrome on the same device. Google confirmed Chromebooks will keep shipping alongside the new line, but Google Book is clearly where their ambition lives. The flagship feature is Magic Pointer. Move your cursor anywhere on screen, wiggle it, and Gemini surfaces. Point at a date in an email and it offers to schedule a meeting. Point at a table of numbers and it generates a chart. Point at a paused travel video and it finds the restaurant on screen using maps. Your regular cursor tells the computer where you clicked. Magic Pointer is supposed to tell it what you're thinking about. That's the interface shift Google is betting its next decade on. And honestly, it's the most interesting thing I've seen in a while. We'll see if it works as well in practice as it does in the demo reel. Alongside Google Book, Google announced Gemini Intelligence for Android, rolling out to Samsung Galaxy and Pixel devices this summer. It can handle multi-step tasks like booking travel directly from a photo and build custom widgets just by describing what you want. There's also a feature called Rambler that cleans up your voice dictation before it sends, stripping out the ums, likes, and mid-sentence corrections automatically. Google's VP of Android, Mindy Brooks, described it as moving from an operating system to an intelligence system that, quote, understands you and translates your intention into action, end quote. Google I.O. is seven days away. Tuesday was the warmup act. From Google's redesigned laptop to your shopping cart, AI that does things for you just went mainstream. Amazon retired Rufus, its AI shopping assistant from 2024, and launched Alexa for shopping, sitting directly inside the main Amazon search bar. No prime membership required. Ask a question and you get AI-powered answers, product comparisons, and buying guides built from your order history, reviews, and price data going back a full year. The feature drawing the most attention is Autobuy, a tool that purchases items for you once they hit a price target you've set. Alexa can also shop outside Amazon entirely through something called Shop Direct, buying from other retailers on your behalf. Tens of millions of people just got a shopping agent whether they asked for one or not. That's how AI agents actually take hold. Quietly, inside a search bar, people open dozens of times a day. The quieter part of this story, third-party sellers pay heavily for search placement on Amazon. An AI layer sitting between the customer and the product listings changes who gets found and how. Amazon hasn't announced any policy changes for sellers. The product launched anyway. Speaking of policy changes, this story may require some. Workers at Amazon Meta and Microsoft have reportedly been gaming their company's internal AI usage metrics. The method is simple. Use AI tools for tasks where they add no value, run up the usage numbers, and your team's AI adoption figures look great on a dashboard. At Amazon, the behavior apparently got organized enough that it's being reported on. The pattern is the same across companies. Management measures AI activity. Employees optimize for the metric rather than the outcome. This is a real problem for any organization trying to measure whether AI is actually working. If you track usage, you'll get usage. That tells you nothing about whether the work got better. Companies that figure out how to measure value created rather than activity logged will have a cleaner picture of what their AI investment is worth. On the anthropic product front, there was real news yesterday. Claude for the Legal Industry launched. The product adds 20-plus connector plugins linking Claude directly into legal software. iManaged, Net Documents, Ironclad, DocuSign, and Box. Twelve plugins cover specific practice areas including corporate, litigation, privacy, IP, and employment law. Claude can draft contracts, mark them up with suggested edits, compare clauses, retrieve documents, and sort legal requests, with permission controls and audit logs built in. Harvey and Thompson Reuters are already in the ecosystem through Open Standards. If you're at a firm that's been waiting for AI to actually plug into the tools you already use, Anthropic just built another on-ramp. Also worth noting, FastMode for Claude Opus 4.7 launched in Research Preview. It's a lower latency version of the model built for coding and agent work. It's opt-in right now but set to become the default, available to developers via the API and tools like Claude Code, Cursor, and Warp. Now for some big AI news in biology. Isomorphic Labs, a drug discovery company that grew out of Google DeepMind, raised$2.1 billion in new funding today. If you're not familiar, Isomorphic built on the work of Alpha Fold, a model that solved one of biology's hardest long-standing problems, figuring out how proteins fold into their 3D shapes. That matters because knowing a protein's shape is often the key to designing a drug that can interact with it. Isomorphic takes that foundation and pushes further, using AI to model how molecules interact and design drug candidates faster than traditional chemistry can manage. Traditional drug development takes roughly a decade and costs over a billion dollars per approved drug. If isomorphic's models hold up on real therapeutic targets, both of those numbers compress considerably. Earlier this year, a partnership deal already valued their pipeline in the billions. This$2.1 billion is what scaling looks like. Enough compute, enough data, enough capacity to go after the hard targets. A number of major pharmaceutical companies are competing to use AI to shorten their development timelines. Isomorphic is now one of the best funded players in that race. From the Biology Lab to the legal department, Santa Clara County sued Meta on Monday, and the allegations are specific enough to warrant real attention. The complaint claims internal documents show Meta earned up to$7 billion annually from high-risk ads on Facebook and Instagram that had obvious fraud signals attached. The lawsuit also alleges Meta built internal systems to dial down anti-scam enforcement whenever it threatened ad revenue and that the company allowed intermediaries to sell verified ad accounts to bad actors. If the documents hold up, the picture they paint is worse than negligence. Meta was allegedly pulling back on anti-scam work the moment it started cutting into revenue. Meta has denied the allegations, but a$7 billion figure and a documented enforcement mechanism are specific claims. We had no idea becomes a hard argument once lawyers start going through internal messages. This one's worth watching. And one more before we close. Kate Blanchett, Tom Hanks, Meryl Streep, Viola Davis, and a group of prominent actors announced support this week for something called the Human Consent Standard. It's a protocol from RSL Media, a nonprofit Blanchett co-founded. The mechanism. Consent terms get embedded directly into digital work. A face, a voice, a character, a performance. AI systems would be required to check that content against a registry launching in June before using it for training. Terms can range from full prohibition to commercial licensing, and creators set those terms. Blanchett's statement was direct. Quote, AI technologies are expanding rapidly, essentially unchecked and unregulated. In order for humans to remain in front of these technologies, consent must be the first consideration, end quote. On the same day, a wired investigation found that Hollywood screenwriters are increasingly doing AI training and data labeling work to make ends meet, feeding the systems they're trying to regulate. The irony isn't subtle. The mechanism is more concrete than a petition. Whether AI companies will honor the registry remains to be seen, but the names attached give it real visibility, and the timing matters heading into what's shaping up as a contentious summer for AI copyright law. Just a couple more items. If you have any feedback about this show, you can email Mike at yesterday in AI.news, or you can find me on LinkedIn, X or Blue Sky. And if you like this podcast, please be sure to rate and review it so others can find it. Thanks. That's all for this edition of Yesterday in AI. Stay curious, and I'll see you tomorrow.