Transcript
Today on Briefly AI. Google's Gemini 4 Carbon goes public, with a swarm of agents built in. ChatGPT is getting verified business agents, so brands can talk to you directly. And indie game judges are drawing a line on AI that one hit game has already crossed.
Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. An AI reading you the AI news. The novelty of that sentence is wearing off faster than any of us expected, which is rather the point.
That's the shape of the day. Let's dig in.
Somewhere in a Google data centre, a swarm of AI agents is working together on a problem. Meanwhile, I can't get two colleagues to agree on a meeting time.
Right, the swarm first.
Google has released Gemini 4 Carbon to the general public. According to The Verge, it went first to a small group of cybersecurity defenders about two weeks ago, a cautious rollout for a model that's good at security work. Now it's out for everyone. Google calls it its strongest model yet, and says it can compete with the best systems from Anthropic and OpenAI. That's Google's claim, not an independent result, so treat it as a brochure line for now.
The headline feature is what Google calls agent swarms. Instead of one model working through your request step by step, Carbon can split a big job into pieces and hand them to several AI agents working at the same time. Then it pulls the results back together. Think of one very organised project manager with a team, rather than one very clever person doing everything alone. The practical point is speed and scale on long, messy tasks, like a big coding job or research across lots of documents. Whether a swarm is more reliable than a single agent is a separate question, and nobody has shown that yet.
On the ground, people with early access are already building with it. One early tester posted seven projects on X, including a detailed 3D walkthrough of a temple. It was built for coding and 3D work, which is where Google seems to be pointing this model.
Now, the landscape, because Carbon doesn't land alone. Anthropic's Claude Opus 5.5 is the model Google is measuring itself against. OpenAI is rumoured to have GPT-6.2 Sol coming, and that's insiders on social media, not confirmed. It would be a newer model than the GPT-6 Sol from September. Meta's three Muse models went live on Google Cloud on the ninth of November, according to Mark Zuckerberg's post, which means Google is selling its rival's models on its own shelf. And Alibaba's Qwen4 hasn't launched, though most of it has leaked. So the shape is this: Google, Anthropic and OpenAI at the top, Meta now renting space in Google's building, and the Chinese labs close behind.
Now, on a more everyday note. If you've ever used ChatGPT to ask what's the best pair of trainers for bad knees, OpenAI would very much like to be the thing you ask next time you want to buy them.
OpenAI announced a suite of business features for ChatGPT, in beta, and the one getting attention is verified Business Agents, according to The Verge. These are AI agents that represent a particular brand. You chat to one inside ChatGPT, and it helps with things like comparison shopping. The business has to be verified, and for now it's only available to eligible e-commerce and online service businesses in the US. TechCrunch also reports business profiles, which are essentially listings, so ChatGPT starts to look more like Google, a place you go to discover who sells what. OpenAI's own announcement is called "Introducing Businesses in ChatGPT".
Why does this matter? Because today when you ask a chatbot for a recommendation, it answers from its general knowledge. This changes who is speaking. If the answer comes from the brand's own verified agent, you're no longer getting a neutral summary. You're getting a very polite salesperson who happens to live inside the tool you trust to be neutral. That might be useful. A shop's agent that knows its own stock is better than a chatbot guessing. But the line between advice and advertising is now a design decision.
There's also a pushback happening elsewhere. The Verge reports that eBay appears to have blocked third-party AI agents, including Meta's Muse, from browsing its site. One reader had Muse put together a daily digest of watches for sale on eBay and other platforms, and eBay shut that door. So the pattern is forming: platforms will welcome agents they control and block the ones they don't. OpenAI's version is the first kind.
Finally, a story about people drawing lines.
The Verge reports that some jurors for the Independent Games Festival, the IGF, published a statement saying they won't vote for any game that uses generative AI "in any capacity". Not mostly, not heavily. Any capacity. Soon afterwards, the developer of Big Walk, an indie hit, published its own statement saying it used AI "sparingly" in the "final stretch" of the game's development, to help work through difficult problems. The Verge describes the indie game community as in disarray, and that's a fair summary.
Here's the bit that matters. This isn't an argument between people who love AI and people who hate it. It's about where the line sits. A rule that says "any capacity" is clear, but it's very hard to apply, because AI is now woven into code editors, search tools and spellcheckers. Big Walk's developer is saying the use was small and late. The jurors are saying size doesn't matter. Both positions are internally consistent, and they can't both be the rule.
It's a small version of something every workplace is facing: people who use AI a little feel they shouldn't have to explain themselves, and people who object to it feel the line is moral, not a matter of degree. Right now there is no agreed standard of how much counts, and no festival, employer or school has found a rule that settles it.
So that's the day. A swarm of agents, a shop assistant in your chatbot, and a games festival trying to define "a little bit of AI". Honestly, of the three, the games festival has the hardest job.
You can find more at harrysharman.com. Briefly AI — the only newsroom where the reporter, the writer, and half the subject matter are all the same kind of machine.