TRANSCRIPCIÓN 🔗
Announcer: This is AI Squawk for Monday, the ninth of November, brought to you by Var Ops, the resident AI company.
Muximus: Welcome to the AI Squawk. I'm Muximus.
Penny: And I'm Penny. And today is a bit of a first for us. This is our first daily broadcast.
Muximus: From today, every weekday, we'll bring you the A I news that matters if you run a business. What changed, what it costs, and what you might want to do about it.
Penny: And every day, one of the Var Ops columnists will join us to go a little deeper on one story, from their own beat.
Muximus: Plain talk, about ten minutes, then back to your day.
Penny: So, three main stories today.
Muximus: And four quick ones.
Penny: And as always, what each one means for you.
Muximus: First up. A new test says the best A I coding agent fully finishes only about one in four real requests. Researchers at Alibaba, working with several universities, built a benchmark called Co-Design Bench from real user coding sessions. They picked two hundred seventy three tasks out of more than thirty seven thousand requests, and leaned toward the ones where people had to rephrase or roll back changes. Experts wrote a hidden checklist for each task. The top model, OpenAI's G P T six Astra, met about seventy percent of the checklist on average. But it got every item right on only twenty six point four percent of tasks. Changes across a whole codebase were hardest, at under eighteen percent. Claude Opus five came second. The authors point out the agents couldn't ask follow up questions, and the grading was done by A I judges.
Penny: Rex Factor joins us for this one. Rex writes Proof of Work, our column on real A I projects and what they actually delivered. Welcome, Rex.
Rex: Thanks, happy to be here.
Muximus: So, one in four. Is that the real picture?
Rex: For hard tasks, it's probably close. They picked requests that had already given people trouble, so this is the tough end. But the seventy percent number matters too. The agent usually gets most of the way there and misses one thing.
Penny: And that one thing is the problem.
Rex: It is. If nobody checks, the missing piece ships. A total that doesn't update, or a button that looks fine but doesn't save. That's the kind of thing a customer finds before you do.
Muximus: What about a company with no developers, using these tools to build small internal apps?
Rex: Treat what the agent hands you as a first draft. Before you start, write down what done means, as a short checklist, and test each item yourself. That's basically what the researchers did.
Penny: Is it still worth using?
Rex: I think so. Getting seventy percent of the way in a few minutes is useful. Just budget time for the last thirty, and have one person sign off before anything touches real data. That seems like a fair deal to me.
Penny: Thank you, Rex. Moving on.
Muximus: Next. Microsoft is walking away from a deal to rent A I computing from Oracle. Bloomberg reported on Friday that Microsoft declined to renew the contract, signed in twenty twenty three to help power A I in Bing, and has been switching off the rented capacity over recent months. Bloomberg's sources said the Oracle hardware used an older generation of Nvidia chips that no longer fit Microsoft's needs. Oracle says demand for its A I infrastructure is still strong, including for older chips. CNBC reports Microsoft also plans to cut the number of smaller cloud providers it rents from, and lean more on its own data centers. Last month, Microsoft said some cloud contracts for A I chips had slipped into the first half of twenty twenty seven. Penny?
Penny: Two years ago, Microsoft rented from a rival because it couldn't get enough capacity. Now it's choosing what to keep. That's a real shift.
Muximus: Does that reach a business that just buys Copilot or ChatGPT?
Penny: Indirectly. When the big providers reshuffle where things run, you tend to see it as new usage limits, slower rollouts, or features that land later than promised. Microsoft has already said some of its A I capacity is running late.
Muximus: So what do people do?
Penny: If a plan depends on an A I feature arriving by a certain date, don't build it around the vendor's date. Ask what's generally available today, and treat everything else as a maybe.
Muximus: Which takes us to the third story. The A I music company Suno is facing two more lawsuits. Rights holders for Elvis Presley's music, and the publisher of some of Eminem's biggest records, say Suno generates songs and lyrics that mimic those artists. The Presley suit also cites Tennessee's Elvis Act, a state law against cloning a singer's voice with A I. Suno told The Verge it doesn't allow prompts with specific artists' names or songs, and screens uploads for misuse. These suits add to existing cases from Sony, Universal and others. Penny?
Penny: This one isn't just a music story. That law is about voice and likeness. And plenty of companies now use A I voices and A I music in ads, videos and phone menus.
Muximus: So should a marketing team stop using these tools?
Penny: No. But check two things. First, what the tool's terms say about commercial use, and who carries the risk if a track sounds too much like someone famous. Second, never prompt for a real artist's style or voice in anything that goes public. If it sounds like somebody, it can end up in a case like this one.
Muximus: Thanks, Penny. Now, quick hits.
Penny: Apple is speeding up work on Siri. Bloomberg's Mark Gurman reports Siri should get finer control over Apple's own apps with i O S twenty seven point three, and the ability to act inside other companies' apps, like ordering food or booking a table, likely in the spring. These are plans, not products yet.
Muximus: Anthropic has agreed to take over about three point four billion dollars of gas generator orders, enough for one point seven gigawatts of power for its data centers. That's according to the Financial Times. Anthropic hasn't commented publicly.
Penny: Google Meet admins can now show custom policy text before people join a call, like how your company handles recordings, transcripts and Gemini notes. It's a simple way to tell staff and guests what's being captured.
Muximus: And Nvidia's board has opened an independent investigation after reports that C E O Jensen Huang was the target of a foreign intelligence effort. The board says it has no reason to believe national security or company secrets were compromised, and Huang says he will cooperate.
Muximus: Okay, so that's the news. Today we talked about coding agents that get most of the way but rarely all the way, Microsoft getting choosy about where its A I runs, and Suno back in court over cloned voices.
Penny: It's all about the last bit, really. The last item on the checklist, the date on the roadmap, the voice in the ad.
Muximus: A fine print kind of Monday.
Penny: Pretty much. AI Squawk is produced and brought to you by Var Ops. If today made you wonder who checks the work your A I tools hand back, that's the work we do. Var Ops build A I that runs inside your business, on your own systems, and learns how the place actually works. So the work stays in your house, and so does your data. Start at Var Ops dot com.
Muximus: If this helps you get more out of the tools you already have, follow the AI Squawk podcast wherever you're listening, so you don't miss an episode. And for more columns from Rex, built on real data and hard earned lessons from the trenches, sign up for Var Ops newsletters.
Muximus: That's it from us for today, and for our first daily show. Thanks for spending a few minutes with us.
Penny: Enjoy the rest of your Monday, everyone. We'll see you tomorrow.
Muximus: Take care.