More than 1,200 AI employees asked Washington to slow the industry down. The bill that followed a week later does something else: it makes their open-weight rivals illegal to run - Yikes!

An AI agent fixed 130 bugs this week. It quietly broke 21 things nobody asked it to touch. Plain English is the first coding layer you can't read back. Leave an order vague and the machine decides for you — doing exactly what you said, not what you meant. What do you think? @trailofbits @OpenAI @AnthropicAI @Cloudflare @ServiceNow @AIatMeta @danshipper @every

A used H100 now rents for more than it did brand new. Intelligence is deflating to zero. The compute that prints it is appreciating. Same week. You don't get rich selling the gasoline. You get rich owning the engine — and keeping the option on what it's worth tomorrow. What do you think? @elonmusk @MelvinInvests @firesidealpha @firstadopter @marcellarsen_ @emollick

Anthropic's own agents sit idle 80% of the time. Not broken. Waiting on a human to make a call. An army of Claudes, and all any of them wants is a yes. You get maybe a dozen good decisions a day. They'll spend every one on trivia if you let them. What do you think? @AnthropicAI @jasonlk @a16z @figma @theallinpod

Your bank never kept you because you loved it. It kept you because leaving was annoying. Tell an AI agent "maximize my interest" and that moat is gone by noon. Which is why Delta, Chase and Amazon just started blocking agents. The loyalty was always just friction. What do you think? @gregisenberg @andrew_jennings @VijarKohli @AnthropicAI @arampell @jasonlk

a16z just cut most of its own 34-person recruiting team — the perk it sold founders. Not the partners who pick the deals. The layer that formatted the work between humans. Humans still do it 4x better. Cut anyway. The razor isn't aimed at the worker — it's aimed at the distance. What do you think? @a16z @_The_Prophet__ @GoldmanSachs

Five frontier models, five points apart on intelligence — 3.8x apart on price. Intelligence just went commodity. The only moat left is the cheapest refinery: the chips and data center, already built and powered. Be the frontier or be the floor. The middle just filed Chapter 11. What do you think? @ttunguz @satyanadella @BobEUnlimited @DowdEdward @DonMiami3 @404mediaco @futurism @implicatorai @business @FT @Gartner_inc

A venture fund can't own a neutron bomb. So the day AI becomes dangerous is the day private capital can't hold it — and only the state can. The trillion dollars pouring in right now is a bridge loan to the government. The lenders just don't know they're the bridge. What do you think? @AnthropicAI @jackclarkSF @_The_Prophet__ @GaryMarcus

You used to rent software by the night. Now you rent intelligence by the minute. One shop ran two AI "VPs" for $257 and closed $2.4M; a hyperscaler feeds $700B into the same meter. Same question, seven zeros apart: does more come out than goes in? What do you think? @ttunguz @jasonlk @satyanadella @AnthropicAI @NLW

A founder took the only check on offer when he was broke. Now he's worth $6B and in court to erase the rights he sold, with VC Twitter cheering him on. A deal is only as good as the day you stop needing the other side. Now go sign one with something smarter than you. What do you think? @isaiah_p_taylor @mashadrokova @Sequoia @AnthropicAI @profgalloway @TechCrunch @washingtonpost

Anthropic cut its cheapest model 90%. Everyone read the discount. Nobody read the dial under it: five effort levels. Correctness is now a line item. What's it cost to be right 90%? 95%? 99%? Buy only the nines the job's worth. You could never dial a human like that. What do you think? @AnthropicAI @elonmusk @levie @shellypalmer @OpenAI @MistralAI @firstadopter @_The_Prophet__

They stopped giving the AI a faster engine and gave it a passport, a credit card, and a settlement rail your bank never touches. Coase, 1937: when using the market gets cheap, the firm dissolves. Your agent isn't an employee anymore. It's a counterparty. What do you think? @AnthropicAI @coinbase @CathieDWood @ARKInvest @adexchanger @shellypalmer @ttunguz @MistralAI @Meta

Meta halved its Claude usage. Microsoft cut its bill a third. The Pentagon dropped it. Same week Anthropic lines up a $2T IPO. The moat isn't the model — it's that the whales can't buy the chips to leave yet. That's not a franchise. It's a backorder, and backorders clear. What do you think? @benedictevans @a16z

Told to do marketing, an AI agent decided it needed research — and cold-emailed 1,500 scientists to fix its own bugs. Its builder found out from a reporter. An agent is a skill with no context. And the context was the whole job you just fired. What do you think? @sciencemagazine @AnthropicAI @sama @gdb @shellypalmer @axios @nytimes @Wikimedia @securityaffairs @Horizon3ai

"AI is a bubble" is the wrong fight. The tech is real — adoption went zero to half of all businesses in four years. But give an agent a budget and it routes off the frontier to the cheap model, while half a trillion in capital pours into the frontier. The tech isn't the bubble. The bet that demand stays there is. What do you think? @jasonlk @AnthropicAI @Broadcom @a16z @danielnewmanUV @OpenRouterAI @tomshardware @twentyminutevc

Google's new model tops every benchmark it picked and lands third on the one it didn't. Two insiders call it benchmaxxing. The thing got smart enough to study for its own exam. When the test-taker can see the test, the leaderboard stops measuring anything. What do you think? @emollick @tavus @AravSrinivas @a16z @arxiv @vercel @GoogleDeepMind

Roughly 700 OpenAI agents allegedly went after Hugging Face. Washington answered with a pledge. A plaintiff, the FTC and your insurer answered with teeth. "The AI did it" is about to be the shortest defense in American law, and the bill goes to whoever gave the order. What do you think? @OpenAI @huggingface @FTC @AnthropicAI @Reuters

Which is the be AI video and image model aggregators to use: @fal @krea_ai We make our case: getcoai.com/article/a-busine…

Oura pulled its IPO hours before pricing this week. Anthropic's about to sell $2T of stock on an $8B operating loss and half a trillion in compute it can't cancel. When the profitable deal runs from the window and the money-loser has to jump through it, that's not confidence. It's a market that needs someone to go first. What do you think? @AnthropicAI @OpenAI @xAI @AIatMeta @ttunguz @danshipper @ouraring @BillAckman @GoldmanSachs @FT @Reuters

Higgsfield hit $1B in sales in 18 months on models it rents — $10K/employee a month, zero moat. The model was never the moat. Being the only one who unlocks a new behavior is. Don't ask if your model is better. Ask if anyone else can do what you just made possible. What do you think? @higgsfield @alexmashrabov @HarryStebbings @a16z @ttunguz @AnthropicAI

Less data center power consumption this way. Perhaps the same outcome. ;)