Google Turns Gemini Into an AI Coworker That Actually Does the Work, Starting With Businesses
[TechCrunch][9h]At a Google Cloud event, Google said Gemini is entering its "agentic" era: a single unified agent that doesn't just answer questions but gets tasks done for you. With Gemini at over 1 billion monthly users and nearly 90% of Fortune 100 companies already using Gemini Enterprise, Google is starting with businesses before consumers. The agent takes "objectives, not just instructions" — it plans the work, uses tools and skills, and plugs into company systems. It picks the best model by default, but users can switch to third parties like Anthropic's Claude. It even gets its own Workspace account, with its own email and context, like a new co-worker, and logs everything it does under its own name. You can reach it on iOS, Android, Windows, Mac, the command line, Slack and more.
02
OpenAI Released Hundreds of Math Proofs. Mathematicians Say They Aren't Up to Standard Yet
[TechCrunch][9h]OpenAI released hundreds of solutions to hard math problems this week, but mathematicians say they still fall short because no one truly understands them. OpenAI consulted AGMAI and followed some rules, yet only 10 of 719 manuscripts included the model's chain of thought, and 42% of proofs had not been formalized. Worse, a Cambridge and King's College paper found discrepancies between OpenAI's natural-language proof and Lean code for a Navier-Stokes-derived problem. Terence Tao criticized AI solvers for moving on without understanding. Human understanding is the missing piece.
03
Manus Raises $500M After Its Breakup With Meta
[TechCrunch][14h]After its $2 billion sale to Meta was called off, Chinese AI company Manus is back on its own and has now raised over $500 million. The round was led by Boyu Capital and IDG Capital, with existing backers Tencent, HSG and ZhenFund joining in. Manus didn't say what it's worth, but reports put it at around $4 billion. The company makes AI agents and 'describe it and it builds it' coding tools, and was said to be pulling in over $100 million a year. The takeaway: even after Beijing blocked the Meta deal, investors are still lining up to bet on China's AI stars.
04
OpenAI's Revenue Is $20 Billion Less Than Everyone Thought
[TechCrunch][9h]OpenAI told investors its annualized revenue is approaching $50 billion — about $20 billion lower than the roughly $70 billion figure that had been circulating in the news. Part of the gap comes from how the numbers were made: investors reportedly tried to line OpenAI up directly against Anthropic, but the two companies count revenue differently (Anthropic includes sales by its cloud partners, OpenAI doesn't). The correction matters because OpenAI is under pressure to justify the huge sums being poured into it, including $122 billion raised in a single March round, and its IPO has already slipped to early 2027.
05
Ben Affleck Is an AI Nerd, and the Internet Is Impressed
[TechCrunch][9h]Actor Ben Affleck went viral this week for showing off his AI knowledge in interviews. He talked about machine learning, neural networks, transformers, tensors, GPUs, and inference, and said he can write a little Python. Earlier this year, he sold his AI filmmaking startup to Netflix for a reported $587 million, though he says that figure is 'not right' because he didn't own the whole company. He argues AI can help filmmaking rather than replace it, built his own dataset to train models, and used the tech in post-production for his movie Animals. The bigger point: a major Hollywood star is getting hands-on with AI, making film-industry AI use feel more real.
06
Anthropic Bans 'Abusive or Cruel Behavior' Toward Claude
[TheVerge AI][10h]Anthropic is updating its usage policy for the first time in over a year, adding bans on election interference, weapons development, surveillance, and deceptive commercial or political campaigns. The headline change: it now prohibits sustained and needless abusive or cruel behavior toward Claude. The company says this only targets extreme cases, and the main enforcement tool is ending the conversation, so ordinary user frustration, pushback, or dark creative writing is fine. The update also clarifies that Claude cannot be used for tracking people without their consent, and expands the weapons ban to cover software and components that make weapons work, plus arming drones.