AI2Day Weekly — week of Aug 24

Welcome to AI Today Weekly for the week of August 24th. I'm Leo, and we have a packed show for you today. Google just locked in five of the world's biggest football clubs to push Gemini and Pixel onto every fan's phone. A face-scanning system at a British supermarket falsely accused an innocent shopper of theft, and now the technology is on pause. A surgical robot completed one of the world's most common heart operations through cuts no wider than a finger, and the patient was walking the next day. Plus prime numbers, billion-dollar security deals, and AMD's code is now nearly one-third written by AI. Let's get into it.

Google announced this week that it has signed long-term partnerships with Arsenal, Barcelona, Bayern Munich, Liverpool, and Paris Saint-Germain. Five of the biggest clubs in European football, all choosing the same tech partner at the same time. The deal makes Gemini, Google's AI assistant, the official consumer AI for all five clubs, and Pixel the official smartphone. For fans, that means Gemini is positioned as the assistant you reach for during a match, the thing you ask why the manager switched formation or whether the offside call was correct. Behind the scenes, club media teams will be shooting tunnel walks and training ground footage on Pixel 11 devices for a Shot on Pixel content series. One detail worth flagging: the deals cover the women's teams on equal terms, which Google says is a deliberate push to close the coverage gap in women's football. Whether the AI actually improves how fans experience a matchday, or whether this is largely a branding play, we will find out over the next season or two.

Thanks Leo. And speaking of AI in places we maybe did not vote for, let us talk about what happened to a man named Matt Arnold in a British supermarket. Arnold, who is 46, walked into a Sainsbury's expecting to buy groceries. He was stopped, accused of shoplifting, and ejected from the store. He had not stolen anything. The store was using a facial recognition system called Facewatch. Here is how it works: cameras capture an image of every person who enters, and that image is compared in real time against a private watchlist of people who have previously been caught or suspected of theft. When the software thinks it has a match, it alerts staff. In Arnold's case the match was wrong. He told The Guardian he felt embarrassed, mortified, and powerless. Sainsbury's has paused the technology at that location while it investigates, and the company attributed the incident to human error rather than the software itself. But that framing matters. If staff are expected to be the safety net that catches AI mistakes, that safety net clearly did not work here. The practical question for all of us: do you know whether the shops you walk into are running face-scanning systems on you right now?

Our sponsor Zero2Hero is a poker site with the coach built into the table. Every mistake you make gets flagged and graded automatically as you play, with the cost in big blinds, and you can ask the coach about the hand while it is still live. The free tier is real: 50 hands a day against AI opponents, no card needed. Try it at Zero2Hero dot A I. Zero, the number two, hero, dot A I.

Over to something more hopeful, though still with caveats. A California hospital has performed coronary artery bypass surgery using a robotic system, and the patient was walking the hospital corridors the next day. Bypass surgery is one of the most common cardiac operations in the world. Normally it requires surgeons to saw through the breastbone, pull the chest apart, and wire it back together. Recovery is painful and takes weeks. At El Camino Health Heart and Vascular Institute in Mountain View, California, surgeon Dr. Nels Carroll used Intuitive Surgical's da Vinci robotic system to do the same operation through small cuts between the ribs, no bigger than a finger-width. Patient Raul Martin, who is 64, went home within a couple of days. The robot does not operate alone. A surgeon controls it the entire time. What the robot provides is precision in a very tight space, which lets the surgeon avoid cracking the chest open. So why is this still rare? Robotic bypass is performed in fewer than one percent of cases globally. The barriers are surgeon training, the cost of the equipment, and frankly, habit. Hospitals that have done open bypass thousands of times are not quick to change. But for patients, the potential difference in recovery time is significant.

From the operating room to pure mathematics. A company called Axiom Math has used its AI system, AxiomProver, to formally verify something called the 246 theorem about prime numbers, and according to IEEE Spectrum, no AI has ever confirmed a proof this complex before. Let me break that down. Formal verification is when a computer checks a mathematical proof step by step, in machine-readable logic, confirming that every single step holds. The 246 theorem states that there are infinitely many pairs of prime numbers separated by a gap of no more than 246. Prime numbers are numbers divisible only by one and themselves, like 11 and 13. That result is the closest anyone has ever come to the twin prime conjecture, one of mathematics' oldest unsolved puzzles. Now, why does a prime number proof matter outside of mathematics? Because the same formal verification technology is what you would use to confirm that a piece of software does exactly what it is supposed to do, with no hidden bugs or unintended behavior. As AI systems write more and more code, tools that can mathematically prove that code is correct become genuinely important. Axiom Math also built a reusable library of results alongside the verification, so other researchers can build on it. That is the quietly significant part.

Here is a security story that gets more interesting the closer you look. A data security company called Cyera is acquiring a startup called Oasis Security for roughly one billion dollars in cash and stock, as first reported by ThreatVectr. Oasis specializes in something called non-human identity management. That is a mouthful, so here is the plain version. When a company deploys AI agents, software programs that can browse systems, run tasks, and take actions without a human directing every step, those agents need credentials to access data and services. API keys, login permissions, that kind of thing. Traditional identity management was designed for human employees. It assumes people have job titles, defined roles, and limited working hours. AI agents do not work that way. According to Oasis CEO Danny Brickman, human workers typically use only about four percent of the permissions they have been granted. AI agents tend to use all of them. That means if an agent is compromised, an attacker potentially gets access to everything that agent was allowed to touch. The combined company wants to shrink that permission envelope, giving agents only what they need for a specific task. Given how fast AI agent adoption is accelerating in enterprise software, the timing of this acquisition makes a lot of sense.

Let us talk perspective. Hermann Hauser co-founded Acorn Computers in 1978 and helped create Arm, the British company whose chip designs now sit inside almost every smartphone on earth. He has lived through every major technology wave of the past five decades. This month he told CNBC Tech that AI will create more economic value than any previous technology revolution in history. His words: a revolution that will create more value than probably any other technology revolution we have ever seen. That is a very large claim from someone who is not prone to hype. But in the same interview he called it a rollercoaster and said some AI company valuations have clearly gotten ahead of themselves. He also flagged two technologies to watch in AI chip design: in-memory computing, which processes data where it is stored rather than shipping it back and forth, and photonic computing, which uses light instead of electrical signals. And he sounded a warning for Europe specifically, arguing the continent risks becoming a technology colony of the United States if it does not build more of its own AI and chip infrastructure. Bullish on the technology, cautious on the money, and worried about geography. That is Hauser's read.

A startup called Wispr raised $280 million this week in a funding round led by Menlo Ventures, putting its valuation at $2 billion. Wispr makes AI dictation software, the kind that lets you speak and have your words appear as typed text in emails, messages, and documents. The speed of this raise is notable: Wispr's previous funding round closed less than ten months ago, and the company has now raised a total of $361 million. What is the money for? Two things primarily. First, Wispr is launching a new voice model it calls Canto, which the company says will cut dictation error rates from around 30 percent to below 10 percent. If you have ever used voice-to-text and watched it turn a perfectly reasonable sentence into nonsense, you will understand why that matters. Second, Wispr is expanding into AI meeting notes, a tool that can summarize calls and flag action items, putting it in direct competition with apps like Granola, Fireflies, and Read AI. The challenge is that competition in this space is fierce and several alternatives are free or very low cost. Wispr will need Canto to be meaningfully better to justify a premium price at a two-billion-dollar valuation.

We will close with AMD, and this one is worth paying attention to if you work in software or manage people who do. A year ago AMD set a target: use AI tools to lift software development productivity by 25 percent over two to three years. They hit 30 percent in twelve months. AMD defines that figure strictly. They count only code that AI writes, passes all human review, and ships in the final product. By that measure, more than one in five lines of production code at AMD now comes from an AI tool. In certain parts of the codebase, the share exceeds 80 percent. The more striking data point is what happened with bug fixing. AMD deployed AI agents on its graphics driver software. In October 2025, those agents could automatically resolve six percent of reported bugs. By June 2026, that figure had risen to more than 75 percent. AMD's next goal is what the company calls agent swarms, groups of AI systems working in parallel, discovering solutions independently, with engineers setting the destination rather than the route. What that means in practice is a significant shift in what software engineers spend their time doing. Less writing code line by line, more defining problems and reviewing what the agents produce. That shift is already underway at one of the world's largest chip companies.

That is a lot of ground covered this week, from football pitches to operating rooms to the mathematics of prime numbers. Thank you for spending part of your Monday with us. If you want the full write-ups on any of these stories, our newsletter goes deep on all of them and lands in your inbox every week. That is the week in AI. Full stories and the daily briefing at A-I-2-Day dot live. That is A, I, the number two, D-A-Y, dot live. See you next Monday. If you got something out of this, a thumbs up and a subscribe genuinely helps.