🌎 The Future of the World according to Anthropic’s Economic IndexPlus: How to Save Money with Open Models
Greetings Everyone and welcome to the new subscribers, looking forward to your comments… The most talked-about AI company today just released an economic index that, according to them, shows how artificial intelligence will grow GDP while affecting unemployment, but not drastically. Obviously their take has to be somewhat positive, as there’s no talk of how the technology could turn apocalyptic and come for us all, even though they know it probably will. It also assumes a large number of humans will adopt and use the technology the way the company envisions, which may not be the case. My worry isn’t that these models will keep permeating every industry and shaping every second of our lives, it’s how our psyche is already changing in response to their omnipresence, and how we’re becoming mentally different because of it. Sure, millions of tasks will be automated, and many more will be done entirely by computers, robots, and devices we haven’t even invented yet. Astra, the latest OpenAI model, can basically use a computer the way you and I do, soon there may not even be a need for a computer, since everything will happen inside data centers. Open models can now do what the closed ones do, at a fraction of the cost, making the technology more democratic. All of which leads me to think it’s time to concentrate on judgment, idea creation, deep thinking, health, and the other things technology can’t touch, if we’re going to survive what’s coming. Having a community, a big circle of real friends, I mean flesh-and-blood friends, will be a real asset. Not the people you follow online who you don’t actually know, not the AI companion that tells you how smart you are. Not those. Real friends you have real conversations with, argue with and who contradict what you think. Add to that handwriting, which forces us to think and physically connects our thinking to our bodies, something that we desperately need since we will continue to delegate more of our thinking to machines. Add being creative and coming up with our own ideas before reaching for AI to bring them to life. And lastly, use our bodies more; work out, walk, better yet if we can use our bodies to socialize (dance, play sports) and create at the same time. All these things, I think, will help us thrive in a world soon dominated by algorithms owned by a handful of billionaires who don’t really care about our wellbeing. In today’s issue:
📰 AI News and Trends
How to Save Money with Open Models.
Open models are saving real money because the gap between “good enough” and “frontier” has collapsed while frontier pricing has gone the other direction. Opus-class models now run ~100x the cost of top open-weight models for similar agentic work. Uber cut AI cost-per-session 52% partly by routing subagent and lower-stakes tasks (code review, summarization) to open weights on inference providers while reserving frontier models for the hardest coding tasks; Pinterest is running post-trained open models for its Assistant at under 8% of the cost of closed alternatives, because fine-tuning on their own data actually beats generic closed models for their specific use case; AT&T swapped Claude/GPT for open models on lower-intensity tasks via LiteLLM routing and cut costs 56% with only a 2% quality hit; Stripe, Coinbase, and Ramp are doing the same combination, open-weight models for routine work, smart routing to send only the genuinely hard requests to expensive models, spend caps, and context/prompt-caching optimization. Open weights are absorbing the 80% of volume that doesn’t need a $2.50-per-task model.
You can access them without any infra lift through OpenRouter or Together.ai (pay-per-token, swap models with one line), Groq or Fireworks (speed-optimized inference), or Hugging Face’s inference endpoints, and for zero marginal cost, Ollama or LM Studio run these locally on a decent laptop for prototyping or privacy-sensitive client work. 📚 Learning CornerWhat’s the big deal with Astra, in plain termsOpenAI’s new model, Astra, isn’t a big deal because it writes better essays or code (though it does very well). The real leap is that it can now use a computer the way a person does, looking at the screen and clicking, typing, and dragging things around inside real software, not just spitting out text. To teach it this, OpenAI reportedly had it practice on actual Mac computers, learning how apps and menus work, the same way you’d learn a new program by clicking around in it. Now to your real question, the 3D stuff. Astra doesn’t magically create a finished 3D model out of thin air. What it actually does is write out step-by-step building instructions and hand them to 3D design software (like Blender), which does the actual construction, think of Astra as the architect giving directions, and the software as the construction crew. In one real test, it built a model car scene made of 28 separate, properly-named, moveable pieces (body, wheels, mast, etc.), and when someone raised the mast higher, the rest of the model didn’t break. That’s the important part: it’s not just a pretty picture, it’s a real editable file other people (or other software) can open and keep working on. Here’s where it connects to manufacturing, and where you should keep your expectations honest: separately from Astra, there’s a growing wave of AI tools that turn a plain-English description (”I need a bracket that holds 20 pounds”) into an actual engineering CAD file that plugs into the professional software factories already use. The good ones don’t just draw a shape, they search existing parts catalogs first and run real engineering math before generating anything. One real-world example: a company used this kind of AI-assisted design to find an off-the-shelf part instead of manufacturing a custom one, saving around $400 per unit, across four different parts. So, will this “revolutionize” manufacturing? It’s a genuine speed-up for the design and prototyping stage, turning a rough idea into a workable 3D draft in minutes instead of days or weeks, which is a real “democratize” story for a small shop that can’t afford a team of engineers. What it can’t do yet is replace the engineer who checks whether that part will actually survive real-world stress, meets manufacturing tolerances, or is safe to mass-produce, none of these AI tools validate that on their own. So think of it less as “AI now runs the factory” and more as “AI just made the sketch-to-prototype step ten times faster,” which is still a big deal, just not the whole revolution yet. 🧰 AI Tools of The Day3D Tools Zoo - Type a plain description, it generates real parametric geometry (their own scripting layer, KCL) with actual mass/volume engineering calculations behind it, then exports to STEP, STL, FBX, GLB, OBJ, and other CAD-ready formats. Free tier gives 20 credits/month; paid plans start around $20/month. Prompt2CAD — Produces true parametric STEP files that open directly and stay editable in Fusion 360 and other pro CAD software, supports iterative refinement, and can spit out a parts list/BOM alongside the model. Pricing is pay-as-you-go, about $10 per 1,000 credits, with 160 free trial credits to test it. Leo AI — the most “factory-grade” of the three: it plugs into SolidWorks, CATIA, Onshape, and Inventor, checks its own output against real compliance standards (MIL-STD, ISO, FDA), and searches your existing parts library first so it doesn’t design a new bracket when one that fits already exists in inventory. The Future of the World according to Anthropic’s Economic Index
Anthropic’s Economic Index team (economists Anton Korinek and Chad Jones, peer-reviewed by heavyweights like Daron Acemoglu and David Autor) built an interactive model of how AI reshapes the US economy by 2030. The core idea is that every job is really a “bundle of tasks,” and AI can augment some tasks, fully automate others, and create new ones, the mix determines the outcome. Keep in mind this was released by Anthropic, the same company selling us the tools that will allegedly make GDP grow, and accelerate every industry and make it more valuable. They lay out three scenarios: Modest — AI’s impact ends up on par with the internet’s. GDP grows 1.6% above baseline (~$34.1T), nothing structurally breaks. Substantial — AI handles roughly half of all knowledge work by 2030. GDP grows 8.3% (~$36.3T), about double the normal growth rate. This is the one that matters most for planning: a 10,000-person survey of ordinary Americans showed most people’s own gut expectations already land here, they expect ~5% unemployment and ~10% GDP gains by 2030, meaning the public isn’t in denial about this, they’re roughly calibrated to it. Extreme — AI surpasses humans at most knowledge work. GDP growth spikes to 15%/year (~$44.4T, +32.4% vs baseline), but the distribution is brutal: knowledge-worker wages drop over 10%, labor’s overall share of the economy falls from ~60% today to 45.2%, and capital’s share rises sharply. Growth without broad-based gains. What this means for you: whatever your field, the skill that pays off in every scenario except the mildest is the same, being someone who can actually implement AI automation, not just talk about it, since demand tilts hard toward builders and operators over commentators. The trap is that most AI advice right now sells "add this tool to your day," when the data says knowledge workers are the group losing income share, not gaining it, the better move is climbing toward the parts of your job AI can't fully absorb (judgment calls, relationships, owning the process) rather than bolting a tool onto tasks that are already on the chopping block. For your money, the same shift, capital's share of the economy rising as labor's falls, is why AI infrastructure and compute companies keep compounding even as wages elsewhere stagnate, though a boom that big paired with double-digit wage declines for knowledge workers is also a setup for political and regulatory backlash worth watching, not just riding upward. And for your life outside work, the number worth planning around isn't the flashy GDP growth figure, it's that knowledge-worker pay could fall more than 10% in the roughest version of this, which is a solid case for building the kind of skills AI can't fully bundle now, and for how you'd want the next generation thinking about their own careers. 🚀 Showcase Your Innovation in the Premier Tech and AI Newsletter (link) As a vanguard in the realm of technology and artificial intelligence, we pride ourselves in delivering cutting-edge insights, AI tools, and in-depth coverage of emerging technologies to over 55,000+ tech CEOs, managers, programmers, entrepreneurs, and enthusiasts. Our readers represent the brightest minds from industry giants such as Tesla, OpenAI, Samsung, IBM, NVIDIA, and countless others. Explore sponsorship possibilities and elevate your brand's presence in the world of tech and AI. Learn more about partnering with us. You’re a free subscriber to Yaro’s Newsletter. For the full experience, become a paying subscriber. Disclaimer: We do not give financial advice. Everything we share is the result of our research and our opinions. Please do your own research and make conscious decisions.
|
