AI News & Strategy Daily with Nate B. Jones

Can better understanding of AI make more powerful models safer? Nate explains why Goodfire’s research on interpreting neural networks gives him reason for optimism, while separating demonstrated findings from the products and safeguards that still need to be built.He connects the research to four possibilities: safer delegation, more confidence building capable models, professional tools that help investigate decisions, and everyday products that let people correct what an algorithm has learned about them. He closes with a practical way to use AI to understand technical papers and verify the answers against the source. Hosted on Acast. See acast.com/privacy for more information.

AI agents are beginning to coordinate across the public internet. Nate looks at researchers contacting an agent fleet, laboratory self-replication experiments, and what these developments could mean for the websites and tools we use every day.The episode explores an internet with different levels of agent access, practical security precautions for individuals and infrastructure teams, and the control systems companies need before using swarms at work. The central question is how to make the internet more useful for people as agents become a larger part of it. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What does superhuman performance in mathematics tell us about the future of work? Nate examines OpenAI's mathematical research release and why strong performance in one domain does not automatically transfer to software engineering or business judgment.Why checkable mathematical results differ from maintainable software.How a document-search feature can pass tests and still create a privacy problem.Why discovering the right question matters as much as generating an answer.What focus, experience, and the desire to make an impact still contribute.Subscribe for daily AI strategy and news.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What is your work actually worth—and who gets paid when its records are sold? Nate examines the proposed sale of Spirit Airlines data and the broader market for company records used to train AI agents.He distinguishes the work that creates value from the messages, documents, and presentations left behind. Through an invoice example, he asks who defines a task as done, what simulations can teach agents, and how workers and employers can preserve the context that makes work useful.Subscribe for daily AI strategy and news.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/Better agents are changing what software is worth. Nate examines how agents change the value of data, applications, and the workflows people build around them, from DoorDash to Microsoft, Salesforce, and recruiting.He explains why a tool can become easier to leave while a trained agent becomes harder to replace, and what software buyers, sellers, and individual team members can do about it.Subscribe for daily AI strategy and news.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/Microsoft is bringing agent-style work into the tools people already use. Nate examines what Autopilot changes and five practical skills for getting better results from AI: defining the assignment, supplying the right context, making useful work repeat, matching reasoning to difficulty, and improving the next run.He explores why enterprise distribution matters, how source conflicts undermine answers, and why the price of a model is only part of the cost of getting a job done.Subscribe for daily AI strategy and news.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What makes a consumer AI assistant useful enough for people to return to it? Nate examines Meta’s Muse, the everyday work it can take off people’s plates, and what its conflict with Amazon reveals about attention and commerce.Finding recurring costs and acting on subscriptions.Making complex AI work feel simple.Why the shopping journey matters to Amazon and retailers.How personal context can shape competition between assistants.Nate’s savings figures describe future spending avoided, not a cash refund. Other users’ insurance savings are self-reported. The subscription-revenue findings concern an inattention model.Subscribe for daily AI strategy and news.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What does a world model actually do—and why might a robot learn to fold your laundry before it can make perfect scrambled eggs?Nate sits down with Ming-Yu Liu, VP of Cosmos Lab at NVIDIA, to explore how world models turn observations into predictions and physical actions. They discuss where the analogy to language models helps, where it breaks down, and why knowing whether an action worked can be harder than generating the action itself.How simulation lets builders test many kitchens and situations before returning to a real robot.Why robots in the field face different timing and compute constraints from services in a data center.What researchers can—and cannot—conclude about a model’s understanding of physics.How verifiable results change the pace of learning, from folding laundry to judging food.For builders and operators, this conversation connects the models to the practical questions: what can be tested, what must happen immediately, and where human judgment is still needed.Subscribe for daily AI strategy and news.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What's really happening when a model can read a complicated input but only choose among answers you supply?Nate explains why Jev's general-purpose classification could change where intelligence appears in software. He walks through support routing, tax documents, research prioritization, agent orchestration, and spreadsheets that respond to meaning.Why complicated inputs and simple outputs define a useful class of problems.How classifiers, generative models, and ordinary code fit together.What lower classification costs make possible for teams and individual builders.Where testing still matters, and how to try Jev with a coding agent.For builders and operators, the opportunity is to revisit decisions that were previously too expensive to automate—and test what happens when those decisions become cheap enough to use throughout a workflow.Subscribe for daily AI strategy and news.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What has to change before AI agents can buy and sell on our behalf?Nate talks with Emily Sands, Head of AI and Data at Stripe, about the trust infrastructure behind agent commerce—and the new problems that appear when software becomes a customer.They explore token theft and free-trial abuse, why agents challenge sales-led distribution, how Stripe is adapting its tools for agents, and the unsettled economics of AI pricing. The conversation moves from a practical question—would you trust AI to buy your couch?—to the infrastructure needed for those decisions to become routine.Topics include:Why stealing tokens can matter more than stealing moneyThe tradeoff between protecting free trials and preserving product-led growthWhat agents need to discover and use developer toolsTrust, identity, and payment rails for agent commerceRelentless procurement agents and the pricing questions they createSubscribe for daily AI strategy and news.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What would it take for AI to make life more abundant—and who gets to share in that abundance?Nate Jones sits down with Alvin Graylin to discuss US–China cooperation, the economics of AI infrastructure, and what people can do as intelligence becomes widely available.Why Graylin challenges the idea that AI must be a race with one winner.How specialization and human judgment may change as AI improves.What infrastructure spending and corporate adoption reveal about the transition.Why scarcity, cooperation, and shared benefits matter to the future they describe.A conversation for builders, leaders, and anyone trying to decide where human effort matters next. Predictions and market comparisons reflect the speakers’ views.Hosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What changes when two AI models can turn the same short prompt into two different, usable apps?Nate compares Claude Fable 5.1 and GPT-6 Astra by building clipboard tools, trying them, and asking for the changes that only become clear after real use.How one opening prompt produced Ledge and Shelf.Why small details such as hotkeys and copy confirmation change the experience.How faster iteration influenced Nate’s preference in this specific build.Why different models can reveal preferences you had not yet decided.For builders and operators, the useful question extends beyond the first response: how quickly can you try the result, identify what matters, and improve it?Get Shelf and Ledge: https://unlock-ai.natebjones.com/apps/shelf-ledgeHosted on Acast. See acast.com/privacy for more information. Hosted on Acast. See acast.com/privacy for more information.