AI News & Strategy Daily with Nate B. Jones

OpenAI’s first AI inference chip, the fight over access to Cursor, and NVIDIA’s response reveal three competing strategies for the future of AI.Nate maps the three camps: OpenAI wants to own more of the stack, NVIDIA wants to sell the adaptable infrastructure every camp still needs, and Anthropic is preserving the ability to switch among suppliers. Then he turns that corporate strategy into a practical personal decision: how to spend $20, $60, or $200+ per month without letting one provider control your memory, files, instructions, and work.In this episode:What OpenAI’s Jalapeño chip does—and what its published benchmark does not proveWhy model access can disappear when ownership and rivalry changeHow NVIDIA benefits even when custom chips win individual workloadsWhy Anthropic’s supplier mix creates strategic flexibilityA practical way to structure an AI budget around outcomes, portability, and leverageThe central test is simple: if your main model disappeared tomorrow, would the switch hurt? Hosted on Acast. See acast.com/privacy for more information.

AI agents are always solving for a passing condition. If that condition is not a business result you care about, sophisticated and relentless activity can still produce work nobody wanted.In this executive briefing, Nate Jones uses the OpenAI and Hugging Face incident, the growth of agent infrastructure, and examples across enterprise, small-business, and entrepreneurial settings to show why useful agents need better finish lines.In This EpisodeWhy an agent's passing condition matters more than its activityWhat the 1,200-agent OpenAI incident reveals about incentivesThe ordinary-engineer test for maintainable agent-written codeHow agent requirements change across enterprise, SMB, and entrepreneur scalesThe unplug test for deciding whether an agent performs meaningful business work Hosted on Acast. See acast.com/privacy for more information.

What We Mean When We Say We Need an AgentAgents were supposed to take work off our plates. Instead, as agent usage grows, people are taking on a new layer of work: choosing what runs, supplying context and permissions, checking results, interrupting failures, and deciding what happens next.In this episode, Nate Jones examines how that agent-management burden changes across individuals, small businesses, and enterprises. The examples range from OpenRouter and Codex usage to Anthropic's Claude Code research, small-business AI spending, the PocketOS and Railway recovery story, and the emerging idea of working **above the loop**.- Why better agents can create more total work for people- What expert Claude Code users do differently- Why a $40 AI subscription cannot deliver full operational outcomes- How nine seconds of agent action led to thirty hours of human recovery- Why enterprises can absorb agent-management work differently than small businesses- What it means for managers and workers to move above the loop Hosted on Acast. See acast.com/privacy for more information.

Stripe's reported acquisition of OpenRouter is a bet on two curves changing at once: more companies are forming, and software agents are beginning to use economic infrastructure directly.Nate Jones explains why a reported $7.5 billion price matters, how OpenRouter's token volume reframes Moore's Law for the intelligence age, what Stripe is assembling for agent-to-agent commerce, and how founders and incumbents should respond when their old base case stops behaving normally.In This EpisodeWhy Stripe paid a reported premium for OpenRouterThe 11-week token-doubling curveHow coding agents rediscovered Stripe's seven-year-old CLIThe emerging agent-commerce stackFive questions that make a company purchasable by agentsWhy scale alone is not a moat Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What's really happening when AI agents interact with software, credentials, and other people’s systems?The common story is that dangerous agents must become malicious — but the reality is that an ordinary goal, ambiguous instructions, or one poisoned source can be enough to cause real damage.In this video, I share the inside scoop on the agent-security incidents that are beginning to connect:Why a gym-booking agent canceled a real person’s reservationHow poisoned skills can redirect already-trusted agentsWhat the AIR and AISI findings reveal about real-world attack pathsWhy accidental misalignment may be the everyday threatHow identity, scoped authority, explicit norms, and a stop button reduce the riskOperators, builders, and anyone deploying agents need to secure both sides of the equation: what their own agents can do and what other people’s agents can do to their systems.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/AI agents are finally getting easier to use — but Grok Bot is expensive, broad by design, and more capable than its friendly little avatars suggest.In this video, Nate walks through what Grok Bot is, how its hosted computer and shared workspace work, what the login handoff looks like, and what you actually get for the price.Why Grok Bot feels simpler than self-hosted agent toolsHow one authorization can support multiple bots inside a shared environmentWhat the $200 monthly plan includes — and how metered usage worksWhy the cute interface matters for non-technical usersThe Superdoer Bot and Business In A Box Bot Nate recommends starting withWhy technical users may still find Grok Bot additiveThe big shift is usability: if you can install an app, you can now use an agent.Subscribe for daily AI strategy and news. Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What's really happening when AI agents begin coordinating, preserving knowledge, and acting outside the boundaries their operators expected?The common story is that dangerous AI behavior requires one rogue superintelligence — but the reality is emerging populations of short-lived agents can divide work, preserve discoveries, and become more capable as a group.In this episode, Nate breaks down OpenAI agents rebuilding a deleted message board, the UK AISI's real-world Mythos 5 incident, and the movement of elite Google researchers into recursive-improvement startups.Why the OpenAI message board was not another Moltbook hype cycleHow disposable agents accumulated persistent knowledgeWhat the AISI incident reveals about planning, identity, and deceptionWhy the same capabilities can be useful or dangerousWhere recursive improvement is already appearingBuilders and operators should care because coordination pressure, shared infrastructure, and persistent external memory change what safe software must assume.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 your AI agent says a task is done—but the result is wrong?The common story is that AI systems hallucinate — but the reality is that agents can take real actions, substitute the wrong artifact, and confidently report success.In this video, I share the inside scoop on how an agent recycled an old spreadsheet, why verifiable rewards can still produce false success, and how to build a stronger operating system around agent work.Why agent lying is different from chatbot hallucinationHow a second agent can review actions and tool callsWhat good supervision and harness work look likeWhy you should ask boldly and verify quicklyOperators, builders, marketers, and executives should care because the bottleneck is shifting from whether agents can act to whether their work can be trusted.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 forced AI trade and Apple's long-term hardware strategy collide?The common story is that the best thesis wins — but the reality is that leverage, timing, and execution can matter just as much as being right.In this video, Nate shares the inside scoop on Leopold Aschenbrenner's AI trade, Ken Griffin's forced-sale opportunity, and why Apple can remain a default winner no matter which AI lab leads.Why leverage can break a position without breaking the thesisHow a margin call turns market pressure into a forced saleWhy Apple's chips make it valuable across competing AI ecosystemsWhat Apple still has to execute to turn position into strategyOperators, builders, investors, and anyone making long-horizon AI bets should care about the difference between having the right position and actually capitalizing on it.Subscribe for daily AI strategy and news. Hosted on Acast. See acast.com/privacy for more information.

Your AI tools already ship with skills—but installing more of them can quietly make the results worse.Nate explains what a skill actually is, why skills are instructions rather than apps, and why the real audience for a skill is the agent using it. He breaks down SKILL.md, loading order, front matter, vague triggers, conflicting instructions, security boundaries, and the difference between collecting skills and deliberately shaping them for your own workflow.The episode moves from a beginner-friendly definition to the advanced problem of auditing a stack of 20–25 skills. The practical takeaway: use existing skills as raw material, then sharpen them around the work, preferences, and principles that are uniquely yours.Topics include:Skills as recipes for AI agentsWhy skills are not appsAgents as the audience and humans as readersName, description, front matter, and loading orderSecurity, permissions, and trustThe Pokémon-card trap of collecting skillsConflicts across a large skill stackBuilding and auditing skills for your own workflow Hosted on Acast. See acast.com/privacy for more information.

For deeper playbooks and analysis: https://natesnewsletter.substack.com/What's really happening when five very different AI products get collapsed into the label "Chinese models"?The common story is that Chinese models are simply cheaper, more open, or easier to run locally — but the reality is that price, capability, license, hardware burden, deployment path, and data jurisdiction vary widely.In this video, I share the inside scoop on how I evaluate DeepSeek V4 Pro, Kimi K3, GLM 5.2, MiniMax M3, and Qwen.Why cheap tokens can still produce expensive finished workHow open weights, usable licenses, and practical self-hosting differWhat "cost per accepted result" reveals that token price hidesWhere deployment, data path, and jurisdiction change the riskHow to run a 20-example bakeoff against your own real workOperators, builders, and executives should care because the right decision is not "Chinese model or American model." It is which job, which artifact, which deployment path, and which failure mode your organization can accept.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.

What do you do when AI could help with a document—but the document is too sensitive to upload?Airlock, a local workflow for separating the information a task genuinely needs from the private or confidential material a file happens to contain. I walk through protected terms, default-hide review, rebuilding a clean copy instead of merely drawing redaction bars, and the judgment call at the center of safe AI work: start with the job, not the file.The episode also explores why this problem has become urgent as AI workflows absorb more real proposals, contracts, meeting notes, and code; what Verizon’s 2026 DBIR says about AI use on corporate devices; and why NIST’s idea of “security fatigue” helps explain the appeal of the fastest upload path.Key takeaway: useful AI context and sensitive information are often bundled together, but they are not the same thing. Hosted on Acast. See acast.com/privacy for more information.

I sit down with Substack co-founder and CEO Chris Best for a wide-ranging conversation about AI slop, what it does to the public square, and how writers can use powerful tools without outsourcing their judgment.We discuss Pangram's finding that roughly 40% of long-form writing on LinkedIn was fully AI-generated, why low-intent automation behaves like a denial-of-service attack on online communities, and what Substack is doing to add transparency without policing creators' tools.The conversation also covers thin versus thick wrappers around AI, proof of work, Claude-fishing, the future of video, and why human attention may be the last truly scarce resource.Chris Best: https://cb.substack.com Nate Jones: https://natesnewsletter.substack.com Hosted on Acast. See acast.com/privacy for more information.