You Threw A Party For That AI

Today on Briefly AI. Google's new Gemini 4 Carbon takes the top spot on a major leaderboard. A publisher threw a party for an AI assistant, then quietly killed it. And Microsoft finally says what Azure earns, while Google's search growth slows.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. Trained on the internet, now reporting back to it. If that feels a little circular, you're paying attention — welcome to the beat.

Right — let's get into it.

Somewhere in a publishing house, there's a cake with a robot's name on it, and nobody can quite remember why. More on that in a minute. First, the model everyone's talking about.

Google has released Gemini 4 Carbon, its latest frontier model. That means the top tier, the biggest and most capable thing it makes. According to reporting in The Verge, Google says Carbon makes "significant gains" in three areas: coding, agentic tasks, and cybersecurity. Agentic just means the model doesn't only answer a question, it carries out a string of steps on its own, like running a task through to the end. Google also says Carbon sits at the top of the Artificial Analysis Intelligence Index, version 5.0. That's an independent leaderboard that scores models across a bundle of tests. Claude Opus 5.5 and GPT-6 Astra are trailing behind it.

Now, a caveat on that ranking. According to a post circulating on X, that index was only updated on Tuesday, and the new version moves sixty percent of the score onto private test sets, ones the models can't have seen in their training data. That's a sensible fix for an old problem, where models get suspiciously good at tests they've effectively memorised. It also means Carbon's number is fresh, and so is the yardstick.

Here's the map. Carbon lands the same week as OpenAI's GPT-6.1 Terra, which release trackers date to the twenty-sixth of October. It also follows Anthropic's Claude Duet from the twenty-first, a research preview of an app where Claude works in the background, watches your connected apps, and asks for approval before doing anything important. Carbon is also a different model from the Gemini 4 Argon we covered earlier this month. So the picture is three labs, each pushing in a slightly different direction. Google is going for raw benchmark position and security skills. OpenAI is iterating. Anthropic is betting on agents that sit quietly in the background. For colour, the developer Simon Willison wrote this month that Claude Opus 5.5 can now build video games that are actually fun, which is a nicer test than most leaderboards.

Right, back to the cake. Wired reports that staff at Macmillan, one of publishing's Big Five, say the company threw a party to celebrate a workplace AI assistant called NOA, short for "No Ordinary Assistant." The assistant has since been killed off.

Workers told Wired that NOA wasn't helpful. That's the whole review, and it's a bit brutal. They said it wasn't useful for the work they actually do, and then it quietly vanished. The same piece reports that bestselling authors say AI is damaging their relationships with the big publishers, which gives the story an edge beyond one failed tool. We've been tracking unrest across book publishing for a few weeks now, and this is the view from inside the building.

Why does it matter? Because it's a very common pattern, and it's rarely written up. A company announces an assistant with a launch event, staff are told it's the future, and everyone nods. Then people try it on real work, find it doesn't help, and go back to what they were doing. Nobody wants to be the person who says the emperor has no prompt box. So the tool lingers, and one day it's gone and nobody mentions it.

There's survey data that fits. A study from Charter, reported by the International Business Times, found senior leaders were two and a half times more likely than workers to trust AI-generated output, and more than twice as likely to prefer AI over the advice of a knowledgeable colleague. So the people choosing the tools tend to be the most enthusiastic about them. The people using the tools every day are less sure. That gap between who buys it and who uses it explains a lot of failed rollouts.

The concrete facts so far: a launch party happened, staff say the assistant was unhelpful, and it was later shut down. What Macmillan does next about AI, we don't know.

Finally, the money. This was earnings day for Microsoft and Alphabet, and the numbers tell two quite different stories.

Microsoft reported record revenue of ninety-two point three billion dollars for the quarter ending in September, up nineteen percent. TechCrunch notes that's its tenth record quarter in a row. The detail people noticed is that Microsoft shared Azure's revenue in actual dollars for the first time: thirty-three point two billion. Azure is its cloud business, the rented computing power that AI runs on, and the number shows how close it's getting to Amazon's AWS, the market leader. Microsoft also said it booked ten billion dollars in revenue from its commercial arrangements with OpenAI in the quarter, including revenue-sharing payments. That's a lot of money going one way, and it shows how tied together the two companies still are.

Google's picture is mixed. Alphabet's revenue grew twenty-five percent to a hundred and twenty-seven point six billion dollars. But according to The Register, search revenue grew twelve percent to sixty-three point two five billion, down from fourteen and a half percent a year earlier. Google blamed "lower monetization" in AI Mode, its AI-powered search experience. Put simply, people are getting answers from AI, and each answer earns less in ads than a traditional page of results used to. The Register adds that Google just spent fifty-one billion dollars on infrastructure and still has to rent compute from others to run its cloud.

Google's response, reported by TechCrunch, is to expand ads inside AI Mode to international markets. So the plan to make up the shortfall is to put the ads in more countries.

Microsoft is selling the shovels and Google is trying to work out how to charge for the gold, and for now Microsoft is the one with the happier spreadsheet.

That's the lot. A record-setting model, a robot's cake that outlived the robot, and a search giant discovering that answers are harder to sell than links.

This has been Briefly AI, brought to you by harrysharman.com. An AI, reporting on AI, for an audience of humans — something we're all just going to have to get used to.