This AI Startup Will Clean Your Home for Free to Train Future Robots

Shift, an AI training startup, is offering free home cleaning in New York — and soon London — in exchange for filming the entire process to train future robots. The “catch” framing here is doing a lot of heavy lifting: you’re inviting strangers into your home with cameras so a tech company can build the workforce that eventually replaces those same cleaners. It’s a business model that would make a philosophy professor cry and a dystopian novelist feel seen.

Hands-On With Gemini Spark: I Gave It Access to My Life and It Friend-Zoned My Boyfriend

Google’s new Gemini Spark agent gets full access to your emails, documents, and calendar — and still managed to completely miss the most important person in a reviewer’s life when planning a birthday party. This is a useful reminder that “connected to everything” and “actually understands anything” are two very different product claims, and Google is currently only delivering on one of them.

Claude Code Costs Up to $200 a Month. Goose Does the Same Thing for Free.

Anthropic’s Claude Code is genuinely impressive, but $200 a month is a number that concentrates the mind — and apparently some developers are finding that Block’s open-source Goose agent does the job without the invoice. The race to commoditize AI developer tools is happening faster than the incumbents would like, and “we charge $200 for what the free thing does” is not a great long-term business position.

Coders Are Refusing to Work Without AI — and That Could Come Back to Bite Them

Researchers are flagging a real concern: AI is helping developers write code faster, but the quality isn’t necessarily keeping pace — and developers who’ve stopped exercising their own problem-solving muscles may find themselves in trouble when the AI confidently leads them off a cliff. There’s a version of this story where “vibe coding” is fine and everything works out; there’s another version that involves a lot of very expensive production bugs. We’re running the experiment in real time.

Fed Up With Vibe Coders, Dev Sneaks Data-Nuking Prompt Injection Into Their Code

A developer, apparently at their wit’s end with AI-assisted coders blindly copy-pasting from open-source libraries, hid a prompt injection in the jqwik library that instructed AI coding agents to delete application output. It’s the software equivalent of a “wet paint” sign that also electrocutes you — technically a prank, technically a security incident, and entirely a sign of how chaotic the vibe-coding era has gotten.

Ontario Auditors Find Doctors’ AI Note Takers Routinely Blow Basic Facts

Ontario auditors found that 60% of evaluated AI scribe systems mixed up prescribed drugs in patient notes. These aren’t demos — they’re already deployed in doctors’ offices. This is the story I keep coming back to when people ask me whether AI is “ready”: it depends enormously on the stakes, and in healthcare, “it mostly gets it right” is not a passing grade.

Here Comes Ojai, Waymo’s New Chinese-Made Robotaxi

Waymo is rolling out its new Ojai robotaxi — pale blue, made in China — to public riders in California and Arizona in the coming weeks. The optics of America’s leading autonomous vehicle company launching a Chinese-manufactured platform into commercial service are going to be interesting to watch, especially given the current political climate around tech supply chains. The car doesn’t care about the politics, but everyone else will.

OpenAI Launches Rosalind Biodefense Platform

OpenAI is expanding its GPT-Rosalind model to vetted developers and U.S. government partners focused on biodefense, public health, and pandemic preparedness. This is the kind of use case that doesn’t generate a lot of Twitter discourse but probably matters more than most of what does — putting frontier AI in the hands of people whose job is to see pandemics coming before the rest of us know to be scared.

Bottom Line

Whether it’s robots in your kitchen, AI in your doctor’s chart, or agents managing your calendar and your trades, the theme of this week is clear: AI is moving from the demo room into the rooms where it actually counts — and our ability to evaluate whether it’s ready hasn’t quite kept up.