01A startup will clean your apartment for free, if it films every scrub for a robot
This week an AI training startup called Shift offered to clean New Yorkers' homes at no charge. The website insists there is no catch. There is a catch. Shift wants footage of the whole job: cleaners scrubbing, vacuuming, dusting, tidying, and washing, recorded from the moment they walk in.
That footage is the actual product. Shift records the cleaners, then uses the video to train robots, according to The Verge. The free cleaning is what the company pays to capture the recording. It announced the offer on social media and says it plans to expand beyond New York to other cities, including London.
The arrangement flips the usual deal. You are not the customer buying a tidy apartment. Your apartment is the set, the cleaner is the performer, and a robot-training model is the buyer. The labor that arrives looking like a favor is closer to a film shoot, and your kitchen is the location.
Shift is one node in a wider supply chain. Paying people to strap on head-mounted cameras and record ordinary chores has become a standard way robot companies harvest training data, Ars Technica reports. Folding laundry, loading a dishwasher, wiping a counter: each is a motion a humanoid robot cannot yet copy without thousands of examples filmed in real rooms. Lab demos do not supply that range. Homes do.
So the data is migrating from controlled facilities into private apartments, and the people generating it are domestic workers wearing cameras on their heads. A Verge reporter, surveying their own flat, conceded the appeal of a free scrub before landing on the trade. The footage of someone cleaning your bathroom does not stay in your bathroom. It becomes part of a dataset meant to teach machines to do the same work later, without the human.
What Shift collects is not anonymized scenery. It is the inside of homes that volunteered for the cleaning, captured in continuous video, owned by the company once recorded. The expansion list starts with London. The offer stays the same everywhere: the service is free, and the recording is the price.
02AI Writes Code Faster, but Researchers Say It Isn't Writing Better Code
Developers are now turning down jobs that bar them from using AI assistants, according to TechCrunch. The refusal signals how deep the dependency runs. It also arrives alongside a warning from researchers: the code those tools produce ships faster, but it is not measurably better.
That gap between speed and quality sits at the center of the concern. Researchers told TechCrunch that AI-generated code can introduce problems that surface later, after the velocity gains have already been booked and the original author has moved on. The cost is deferred, not avoided.
A developer blog post circulating widely draws a historical parallel. It argues AI is repeating frontend development's "lost decade," when layers of tooling and abstraction piled up technical debt that teams spent years unwinding. The post drew 289 points on Hacker News, where the comparison resonated with engineers who lived through that period. The thesis is blunt: convenience now, cleanup later.
The most pointed caution comes from inside the industry that sells the tools. Cognition built Devin, the first and arguably most successful AI coding agent. Its leader, Scott Wu, says the agent was never designed to replace human programmers. A man whose company profits directly from coding agents is telling customers not to hand over the work entirely.
Put together, the three signals point one direction. Teams are trading code quality and developer skill for output speed, and the bill comes due downstream. The dependency makes the trade harder to reverse. Developers who refuse to work without AI may also be losing the ability to evaluate what it produces.
The practical implication lands on engineering teams making staffing and review decisions now. Faster shipping looks like productivity on a dashboard. Whether it holds up depends on who is left to read the code when the agent's output breaks, and whether that person still knows how.
03OpenAI Sends Its Frontier Model Into Biodefense Labs. A New Papal Encyclical Says "Technology Is Never Neutral."
OpenAI is moving its most capable models into the places where failure kills people. The company launched Rosalind Biodefense, expanding what it calls trusted access to GPT-Rosalind for vetted developers and U.S. government partners working on biodefense, public health, and pandemic preparedness. The pitch is straightforward: frontier AI as societal resilience, a defensive shield built before the next outbreak arrives.
The clinical case sits alongside it. Boston Children's Hospital says it uses OpenAI technology to improve patient care, cut operational load, and help diagnose more than 40 rare disease cases. Children who went undiagnosed now have names for what afflicts them. The optimistic story almost writes itself, which is the problem.
Pope Leo XIV's new encyclical takes direct aim at that reflex. Magnifica Humanitas carries a line MIT Technology Review flags for technologists and policymakers: "Technology is never neutral." The document is described as a call to act with courage and solidarity as AI drives what it labels the greatest change of the age. The argument cuts against the idea that a beneficial deployment is self-justifying.
The collision is sharpest exactly where OpenAI is most confident. Biodefense is dual-use by definition. The same models that help vetted partners anticipate a pathogen carry information about how pathogens work, which is why access runs through vetting rather than an open door. OpenAI's own gatekeeping concedes the point the encyclical insists on: even the most helpful deployment carries values someone chose, and someone has to hold them.
Neither side disputes the upside. The rare-disease diagnoses are real, and the biodefense mission targets a genuine threat. What the encyclical contests is whether good outcomes settle the question of who decides, on whose terms, and with what accountability. OpenAI's answer is its vetting process. The encyclical's answer is a moral framework owned by people, not platforms.
The encyclical does not name OpenAI, and OpenAI does not answer the encyclical. They are arguing past each other about the same deployments.
Google showed Gemini Omni and Gemini 3.5 running live at I/O 2026 Google posted nine demo videos of Gemini Omni and Gemini 3.5, the models it unveiled at I/O 2026. The clips cover multimodal input and agentic task completion. blog.google
Groq seeks $650M to fund a shift from chips to inference Groq is raising $650 million in internal funding, Axios reported, as it moves from selling hardware toward AI inference services. The round follows Nvidia's $20 billion arrangement that pulled away parts of the company. techcrunch.com
Glean tripled revenue past $300M by pitching itself as a cost cut Glean's annual revenue crossed $300 million, three times its prior figure. The enterprise search startup now sells itself on trimming AI budgets, even as Microsoft, Google, and others entered the same category. techcrunch.com
XCENA raised $135M to bet that memory, not compute, limits AI South Korean chip startup XCENA closed $135 million at a $570 million valuation. The company argues memory bandwidth, not raw compute, caps AI performance, and builds processors aimed at that constraint. techcrunch.com
Google's Gemini Spark agent planned a party but missed the user's boyfriend Wired tested Gemini Spark, Google's agent that reads a user's emails, documents, and calendar. It planned a birthday party but failed to identify the person closest to the reviewer. wired.com
Adobe's image agent works through conversation rather than one-shot prompts Adobe released a conversational AI image assistant that revises designs through back-and-forth chat. The Verge found it more collaborative than typical text-to-image tools but compared its output to a mediocre design intern. theverge.com
Amazon greenlit an AI-animated show using a character without its creator's consent Amazon is producing an AI-animated series based on The Good Advice Cupcake. BuzzFeed licensed the character to Amazon for AI production; original creator Loryn Brantz said she never agreed and is furious. wired.com
Box's Levie warns executives cut jobs for AI without understanding the work Box founder Aaron Levie said leaders replacing staff with AI often least understand those jobs, calling it "AI psychosis." ClickUp recently cut 22% of its workforce for AI agents, and 2026 tech layoffs already approach the 2025 total. techcrunch.com