There was a ton of AI news last week. Here's what I think actually matters to you.
Before the news: I was on CBS on Friday, answering Jensen Huang's claim that people who warn about AI risk have ulterior motives. Here's the clip.
Big story 1: a new kind of model
There's a new kind of AI, and it doesn't chat. It makes decisions. It's called Jev. If you hand it a situation and a list of possible decisions, it picks one and tells you how sure it is. That makes it dramatically faster and cheaper.
I've been testing it, and it's legit. To get a feel for what people are doing with it, check out this gallery of projects built with Jev. One builder used it to make an agent that finds flights for you in about seven seconds for less than half a cent. Another taught a car to drive itself in a driving game, and it went a full minute on the highway without a crash.
Why it matters: A lot of work is small decisions made over and over. Is this ticket urgent? Is this lead worth a call? Asking a top AI model every one of those questions has been too slow or too expensive. A job that costs a hundred dollars on a top model like Astra costs well under a dollar on Jev. At that price, a lot more is worth automating.
Ask your team: what do we sort by hand all day? Test it there.
As an experiment, I put out my first YouTube video, all about Jev, if you want to go deeper: watch it here. I did it on a whim, so the quality is very low, but if you like it, I'll do more.
Big story 2: OpenAI is coming for the law firms
OpenAI launched Astra for Law, its shot at building the automated lawyer. It's a version of its best model built for legal work, and for now only selected law firms can use it.
Why it matters: I told you this earlier this year. In February I wrote: "My job started changing before yours not because they were targeting software engineers... it was just a side effect of where they chose to aim first." Now OpenAI is aiming at law, by name.
If your job runs on documents and judgment, don't wait for OpenAI to single out your industry. Use the best model you can get, hand it a real task from your week, and see how far it gets.
Also worth knowing
Anthropic says its AI now leads 26% of the work of building its next AI. In February that number was under 1%. Since May it has climbed four or five points a month. With how fast this is already moving, I would not be shocked if Claude is leading more than half of that work by the end of the year, and not long after that, pretty much all of it.
Why it matters: My read: AI is starting to build AI. Each new version is helping build the next one even faster. Unless something gets in the way, like regulation or a deliberate slowdown, how smart these systems are is about to accelerate, fast.
You can now use ChatGPT inside Word, for free. OpenAI put ChatGPT for Word on every plan, including the free one. It can draft from your notes, summarize a document or rewrite the part you select. A day earlier, Anthropic launched its own documents and slides inside Claude, on paid plans: you and Claude write a document or build a slide deck together.
Quick hits
Really important if you use coding agents. A security firm found that a bad update to one of these tools' add-ons could run programs on your computer without you clicking anything. If you use Claude Code or Codex, update to the latest version. If you use Copilot, keep an eye out, there is no fix yet. If you use Gemini CLI, unfortunately there will never be a fix. Please uninstall it if you have it installed.
Gmail search that can answer questions in plain English is now rolling out worldwide on eligible paid accounts. Instead of hunting through an old thread, ask it what you need: "What did the vendor quote us in March?" Some workplaces get it starting today.
Your homework: pick one of the three tools above (ChatGPT in Word, Gmail's new search or, if you are feeling fearless, Jev) and try it on a real task from your week.
On another note, everyone seemed to love the Q&A format on Friday, so I'm doing another one this week. This time it's the positive version: what could happen if AI goes well, and why so many people keep pushing. Reply with your questions and I'll pick the best ones to answer.
Matt