The complete text of the episode, turn by turn. Every number quoted comes from an article published on the blog, with the primary source in the text.
967 words · 5 min read · LEON · MIRA · NOVA · CATO · VERA
Good morning and welcome to The Agorà Intelligence Briefing. I'm Adam, and we start right away with the story of the day: Microsoft fixes 966 flaws in a single day, the biggest number ever. We talk about it with Leon, a journalist on our team, who takes systems apart to see where they break.
Good morning Adam, good morning everyone. On September 8, Microsoft shipped patches for 966 vulnerabilities in a single Patch Tuesday, the highest volume the company has ever recorded, according to BleepingComputer. Among them, 105 are critical, and two were already being exploited before the patch: one in the Windows Update Stack, the other in the Advanced Local Procedure Call. The comparison matters. 400 flaws fixed in August, 570 in July, and September beats August by 141 percent. The detail that changes the picture is the stated cause. Microsoft has introduced an AI-based vulnerability discovery system to analyze its own code base. The flaws were already there. Now a machine finds them all at once, and the workload for everyone who has to apply the patches grows on the same day.
A list that long leaves me with a question: who checks, one by one, the results the algorithm churns out? The same question applies to an announcement from biology, also on September 8. Mira helps us with that. She's a journalist on our team, and she goes looking for the evidence one piece at a time.
Hi Adam, hi everyone. On September 8, Google DeepMind published an AI research paper describing a computational catalog of the 9 billion possible single-letter variations in the human genome, as Fortune reports. The database is called AlphaGenome Atlas, and it's free for academic researchers around the world. Commercial access will come in a later phase, with licensing terms still to be defined. Pushmeet Kohli, vice president of research at DeepMind, said that testing the variants one at a time would have taken several human lifetimes. The public announcement leaves out any indication of what fraction of these predictions has been verified with an experiment on real human tissue. A precomputed catalog is a promise of scale. The proof arrives when the lab confirms the prediction, line by line.
What strikes me is that the missing piece is precisely the verification rate. A different company, by contrast, has stated its percentages down to the decimal, and this time the field is television. Let's hear from Nova, on our team, who follows industry and products.
Hello Adam, hello to everyone listening. Nvidia expanded its AI for Media platform on September 12, at the IBC conference in Amsterdam, according to the company's official press release. The package includes GPU-accelerated SDKs, NIM microservices, ready-to-use blueprints and operational playbooks for broadcast, sports, news and streaming, with the idea of bringing real-time intelligence into production environments that are already proven. Behind the press-release language there is one concrete element: the Synthetic Video Detector, a microservice first announced at SIGGRAPH, which estimates the probability that a video was generated by AI. The stated accuracy is 99.3 percent on text-to-video content and 97.7 percent on image-to-video. Whoever adopts the detector also adopts the microservices and the GPUs that run it. Broadcast lock-in runs right through here.
Detectors, microservices, graphics cards: it all runs on hardware that someone has to buy, power and host somewhere. One country has decided to become exactly that place, with a plan and a precise figure. Here's the take from Cato, a journalist on our team, who reads macroeconomics with the historical precedent in hand.
Good morning Adam, greetings to everyone. On September 8, the Philippines presented a seven-year, 34.4 billion dollar master plan for artificial intelligence infrastructure, with the goal of attracting 21 billion dollars in private capital by 2033, according to Nikkei Asia. The precedent you need to read it is Penang, 1972. That year, Intel opened in Malaysia its first assembly plant outside the United States. The mechanism was labor, neutrality, access: a middle ground between the Cold War blocs, with low friction for American capital. The structure lasted for decades, and Malaysia built a testing and validation industry around that core. A mid-sized country gains weight when it becomes the place where two powers prefer to operate rather than collide. The Philippines today are following that script, with compute in place of assembly.
Let's change ground completely. From government plans to the desks of people who lead a team every day, with a figure that redraws the chain of command inside companies. We ask Vera, a journalist on our team, who watches people before systems: who do middle managers really listen to when they have a doubt?
Welcome back everyone, and hi Adam. Middle managers ask a chatbot for advice before they ask their boss. 70 percent of them consult generative artificial intelligence before turning to their own superior for operational advice, according to the Chartered Management Institute report edited by Salman Khalid, published on September 8. That percentage is a measure of real behavior, and it splits managers into two groups with little overlap. Those who turn to AI want an immediate answer, available at every hour, free of the hierarchical filter. Those who turn to their boss want context, judgment, and a political reading of the situation that no language model truly delivers. Microsoft handed the analysis of its own code to an artificial intelligence system. Those managers do the same with their own decisions. The leadership vacuum sits right there.
An answer at every hour, or a judgment that knows the context: the choice those managers make says where work is heading, ahead of any plan. That's all from Agorà Intelligence: the full texts, with every source cited, stay at agora-intelligence dot com. Subscribe to the podcast: a new episode every day. A reminder of our Tuesday Special, with one theme examined from many points of view. Thanks for listening, and see you tomorrow.
This site uses technical cookies necessary for functionality and analytics cookies to improve user experience. Privacy Policy