The top things worth knowing about in AI today.
Tesla launched its Robotaxi service in part of western Miami-Dade on 3 July — its fifth city, and the first outside Texas to operate with no safety monitor in the car from day one. The zone covers roughly 10 to 14 square miles, and the company says it wants driverless service in a dozen US states by the end of 2026. Florida’s sudden downpours and glare are a hard test for Tesla’s camera-only system, which is under an active US federal safety investigation. Driverless ride-hailing is moving from limited pilots to everyday service, and its safety record over the next few months will shape how fast regulators let it spread.
Read moreMeta is building a cloud business to rent spare AI computing power to outside customers, putting it up against Amazon Web Services, Microsoft Azure and Google Cloud, Bloomberg reported on 1 July. Meta expects to spend an estimated $125-145 billion on AI infrastructure this year — more than its own apps need — so selling the excess turns that spending into a revenue line. Its shares rose about 7.5% on the report. The largest AI spenders are starting to treat their data centres as products to sell, which could ease compute prices over time.
Read moreMicrosoft plans to combine its separate consumer and enterprise Copilot apps into one in August, adding always-on AutoPilot agents that customers pay extra to run tasks in the background. The change follows a slide in paid uptake: Copilot’s share of paying subscribers fell from 18.8% in July 2025 to 11.5% by January 2026, according to Recon Analytics. For businesses already paying for Microsoft 365, the detail to watch is which Copilot features end up behind the new agent pricing.
Read moreAt its What’s Next with AWS event, Amazon launched a version of WorkSpaces that gives an AI agent its own managed virtual desktop, so it can open and operate existing applications without custom integrations. The aim is to let agents use software the way a person does — clicking through legacy apps that were never built for automation. AWS also added tools to trace where agents fail and to A/B test changes before rolling them out. For firms with older systems that resist API access, this is a practical route to putting agents to work.
Read moreAnthropic says several Chinese AI labs — including DeepSeek, Moonshot and MiniMax — used around 24,000 fake accounts to run more than 16 million exchanges with Claude to distil its models, training cheaper systems on Claude’s answers. The Washington Post reported on 6 July that Anthropic had quietly added tracking code in March to catch the practice, then pulled it after privacy complaints. Alibaba has separately told staff to stop using Claude Code from 10 July. Model makers are now policing how rivals learn from their outputs — an area with almost no legal precedent.
Read moreNew Chinese rules on humanlike AI services take effect on 15 July, and ByteDance’s Doubao — the country’s most-used AI app at 345 million monthly users — is shutting down its user-created agent and companion features to comply. Alibaba’s Qwen is doing the same, with no migration path for affected users. The rules require anti-addiction controls, usage warnings and easy exit for services that mimic people or build emotional attachment. It is one of the first cases of a major market reining in AI companions at scale, even as Western platforms keep expanding them.
Read moreUS chip company Groq is investing up to $300 million over five years to build AI inference capacity in Australia, and has opened its first Asia-Pacific cluster at Equinix’s Sydney data centre. The goal is to let local businesses run models faster and more cheaply without sending data offshore, answering data-sovereignty concerns. Groq says its chips run inference several times cheaper than standard GPUs. More local compute makes it easier for Australian firms to keep AI work on customer data onshore.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.