Good morning {{first_name| Artificial intelligence lovers}}. Most AI agent demos live inside sandboxes with fake money and simulated users. This person signed a three-year lease and hired real people.
Andon Labs’ latest experiment has brought AI into its San Francisco retail store with a $100,000 budget and complete autonomy in hiring, operations, and more, a preview of a future where AI replaces the boss long before it replaces the worker.
-
AI agent hires humans, sets up shop in San Francisco
-
OpenAI talks about human competition, and Amazon is on the upside
-
Run Google’s latest AI technology on your phone for free
-
Stanford AI Index: 53% approval, 31% confidence
-
4 new tools for AI, community workflows, and more
Artificial intelligence experience
Rundown: Andon Laboratories only decreased An AI client hired Luna into a real retail space with a budget of $100,000 and a credit card, where the AI set up shop, hired workers, and managed the store as the world’s first AI employer.
-
Andon Labs’ latest experiment was with artificial intelligence Vending machine at Anthropic, with the new contract granting a 3-year lease, a $100,000 budget, and complete autonomy.
-
Luna’s only directive was to make a profit, as the AI created the boutique concept, posted job listings, and handled interviews via Zoom (camera off).
-
proxy runs On Claude Sonnet 4.6 for inference and Gemini 3.1 Flash-Lite Preview for audio, monitoring the store via screenshots from security cameras.
-
When hiring an illustrator, Luna mistakenly selected Afghanistan in TaskRabbit’s drop-down list, and later messed up the staff’s opening weekend schedule.
Why it matters: Real-world customer experiences like this continue to yield the same result: capable in some areas, but laughably dysfunctional in others. But every model upgrade, memory advancement, and proxy feature will help close that gap, with a Luna version that doesn’t make these mistakes likely just a generation or two away.
Rundown: Most teams choose a search provider by running a few test queries and hoping for the best, which is a recipe for hallucinations and unexpected failure. This technical guide is from Ant.com Gives you access to a specific framework for evaluating AI search and retrieval.
-
A four-stage framework for evaluating artificial intelligence research
-
How to create a golden set of queries that predict real-world performance
-
Scales and code to measure accuracy
Obinay
Image source: Louvart/Rundown
Rundown: OpenAI’s CRO Dennis Dresser sender An internal memo describes Anthropic’s $30 billion run rate as “inflated,” describes it as a “one-product company in the platform wars” and points to the Amazon deal as a way to break free from Microsoft’s constraints.
-
memo, published By The Verge, he calls Anthropic’s computing shortage a “strategic mistake,” saying users now face restrictions on access and availability.
-
Dresser said Anthropic’s message is built on “fear” and “limitations,” and that “OpenAI’s positive message will win out over time.”
-
The company also accused the rival of inflating its revenue figures through accounting methods, alleging that it overstated the operating rate by about $8 billion.
-
Hairstyle too Named Microsoft’s deal restricts OAI to enterprise business, citing “amazing” demand for Bedrock since Amazon in February deal.
Why it matters: Either OpenAI uses “internal” memos to strategically leak information to the media, or it’s just really bad at keeping things in-house. Either way, this memo sounds more like an IPO pitch than a strategy update – and with both competitors racing to make their public debuts this year, that’s probably the point.
Artificial intelligence training
Rundown: In this guide you’ll learn how to download Google’s latest AI models and run them on your phone for free. No account, no subscription, no internet needed after setup.
-
Once the form is ready, tap the Settings icon at the top right and enable Reflection. This thinking begins step by step, so the model demonstrates its operation
-
Start chatting now. Everything runs locally on your phone, without any data leaving the device, and the model works even without an internet connection
-
One caveat is that the app does not save chat history. If you close a topic, the conversation is over. However, you can press the + button to see sent messages
Pro Tip: Tap on agent skills instead of AI chat to give the model additional abilities. There’s restaurant roulette, a Wikipedia search, an interactive map, a QR generator, and more.
Submitted by Glenn
Rundown: Join Glean on April 23 to learn why connectors, MCP, and access to real-time applications are not enough. Learn the shortcomings of monolithic approaches and why a permissions-aware context graph creates a stronger foundation for AI and agents in the enterprise.
-
Why standardized approaches fail to achieve quality, depth of context, and token efficiency
-
How to give AI and agents the context needed to think and act reliably
-
What to look for in an enterprise AI platform
Artificial intelligence research
Image source: Stanford High
Rundown: Stanford High Released The 2026 AI Index, which shows the technology has now reached more than half the world’s population faster than a personal computer or the Internet – but with public confidence in AI at record lows and entry-level workers already losing their jobs.
-
Nearly 3/4 of AI experts are optimistic about the technology’s impact on jobs, but only 23% of the public agrees, the largest gap tracked by the report.
-
The United States is building the most AI systems in the world, but ranks only 24th in its actual use at 28.3%, behind Singapore, the United Arab Emirates and most Southeast Asian countries.
-
China nearly erased the US lead in AI standards with Anthropic’s top model leading by 2.7%, while AI researchers moving to the US fell by 89%.
-
Employment of developers between the ages of 22 and 25 has fallen by about 20% since 2024, even as the number of older engineers increases, and company surveys suggest that the planned reductions will accelerate.
Why it matters: These are just a few of the countless interesting statistics in over 400 pages a report. The gap between experts and the public is a timely statistic, given the current anti-AI climate Scary ways. AI insiders see a productivity boom, but ordinary people don’t believe it, and only 31% of Americans trust the government’s ability to manage the changes.
-
📚 Scroll – Transform any knowledge base into a world-class AI experience for employees and customers*
-
⚖️ Harvey Agents – Legal AI agents for memos, diligence reports, and more
-
⚙️ HeyGen CLI – Proxy tool for generating videos via the terminal
Workshop laboratories Announce She will join the Thinking Machines, bringing human-specific AI kits to Mira Moratti’s lab by acquiring them.
apple And it is said building Its first smartglasses with four frame options and an oval camera system, with plans to rival Meta’s Ray-Bans as early as 2027.
Legal AI startup Harvey Fired Agents, autonomous robots that can execute complete legal workflows, including research, memos and slide decks across 13 domains.
Microsoft He is building OpenClaw-style features for 365 Copilot, including 24/7 agents within Office applications, with a possible preview at its Build conference in June.
Softbank Fired A new company backed by NEC, Honda, Sony and five other Japanese companies to build a domestic physical AI model with a parameter of 1T.
In each newsletter we show how the reader is using AI to work smarter, save time or make life easier.
Today’s workflow comes from the reader Bethany M. In Austin, Texas:
“When the family moved to San Francisco, I inherited all the pantry items they had left behind. With my pantry already full, I was suddenly surrounded by unfamiliar flours, grains, and dry goods-and I had no idea how to use them.”
Instead of trying to list everything, I took a picture and asked ChatGPT what I could do. One of my first successes was turning a mixture of flour and grains into a surprisingly delicious loaf of bread in my bread maker.
This small win gave me the confidence to keep going. Now I use ChatGPT to help with all my cooking, like making substitutions, reducing sugar, and adapting recipes. I waste a lot less food and feel more creative in the kitchen.”
How do you use artificial intelligence? Tell us here.
That’s all for today!
Before you go, we’d love to know what you thought of today’s newsletter to help us improve The Rundown experience for you.
Rowan, Joey, Zack, Shubham and Jennifer – the humans behind The Rundown
