Alex Rattray argues that the internet’s plumbing was built for humans and conventional software, not language models, and that today’s Model Context Protocol often buckles under the weight of context, tool sprawl and weak security. The conversation traces a different path: AI agents that write and execute code against typed APIs, with permissions enforced at the API layer rather than bolted onto chat interfaces.

Noah Breyer describes an unusually practical AI stack: an Obsidian vault synced to a home server, cloud coding agents that search and summarize his notes, and a phone-based workflow that turns dead time into deep work. The conversation broadens into a larger argument about probabilistic software, bureaucratic organizations, and why AI is often more useful as a reader and thinking partner than as a writer.

Anthropic’s platform leads sketch how AI infrastructure is shifting from a bare completion endpoint to managed, long-running agents with memory, tools and cloud scaffolding. They argue that the real bottleneck is no longer prompt craft but the dreary work of production infrastructure, and imagine a near future where Claude chooses its own model, spins up sub-agents and compresses work down to an outcome and a budget.

A once-clunky coding tool is recast here as a desktop command center for modern office work, with hosts arguing that the real contest in AI is over agent interfaces that sit between workers and their apps. The conversation moves from theory to practice, showing how Codex and similar tools now draft emails, triage workflows, assemble strategy docs, and turn fragmented digital labor into something closer to managed delegation.

Stripe’s Emily Glassberg Sands sketches an internet where AI agents are becoming buyers, sellers and developers, forcing payments, fraud detection and software tooling to adapt to machine users. The conversation traces the surge in AI-company revenue, the rise of compute theft and free-trial abuse, and the new pricing and checkout systems emerging for an agent economy.

Quora GM Kieran argues that AI can automate the middle of knowledge work, while humans remain essential at the beginning and end: framing problems, exercising taste and polishing what ships. The conversation recasts coding agents as collaborators that free workers to pursue the parts of their jobs that feel most personal, creative and alive.

Logical Intelligence founder Eve argues that energy-based models could make AI more inspectable, efficient and constrained than token-by-token language models. The conversation makes the case for verifiable systems in code, hardware and other safety-critical work where hallucination is not an acceptable trade-off.

Employees at Every describe how personalized AI agents are reshaping office life, from handling household errands and email to sharing skills, filing bugs and collaborating in Slack. The conversation weighs the promise of specialized digital coworkers against the thorny questions of trust, memory, security and workplace etiquette.

Linear CEO Karri Saarinen argues that AI should accelerate execution, not replace the slower work of identifying worthwhile problems. As coding agents proliferate, Linear is positioning itself as the shared context and decision-making layer that guides automated work without surrendering product judgment.

Instagram co-founder Mike Krieger weighs how AI is speeding software development while making product judgment, deletion and durability more important. He and Dan Shipper discuss agent-native design, lean teams and the challenge of building tools that feel flexible without becoming fragile.

Every editor-in-chief Kate Lee traces her path from literary agencies and Stripe Press to building a small, AI-literate newsroom. She explains how style guides, agent browsers, and shared editorial systems can raise the floor without surrendering judgment or taste.

Every’s team introduces Proof, a free and open-source collaborative editor built for humans and AI agents to draft, comment, revise and trace the origins of text together. The conversation weighs provenance, creative ownership and the emerging divide between writing meant to be read and writing meant to be processed by machines.

Bolton and Watt’s founders make the case for “AI-durable” companies: operationally complex businesses such as med spas and funeral homes whose core services remain human even as their systems get smarter. They describe using AI to speed research, sharpen customer discovery and raise expectations for teams without mistaking generated output for good work.

A 17-year-old Alpha School student describes an AI-shaped education built on self-paced coursework, human guides and entrepreneurial projects. She also weighs her generation’s anxiety about automation against a conviction that technology should make more room for connection, judgment and creativity.

OpenAI’s Codex leaders make the case for a dedicated agent command center, where programming moves beyond IDEs and terminal workflows into persistent, multimodal collaboration. As coding models accelerate, they argue that verifying the flood of generated code—not producing it—will become the real bottleneck.

Atlas engineers Ben and Darren trace the evolution from Netscape and Chrome to an agentic browser built around ChatGPT, where web pages become both workspace and collaborator. They weigh automation against human curiosity, and explain how AI coding tools are changing the speed of software decisions as much as software production.

Every’s consulting chief Natalia shares what she has learned from more than 100 companies trying to turn AI enthusiasm into durable operational change. The conversation follows a private-equity firm’s tailored investment-memo workflow and an AI project manager that combs through workplace data to run client engagements.

An entrepreneur describes using Claude Code to build bespoke tools for relationship counseling, email triage, personal styling and school logistics, often in hours rather than months. The conversation weighs AI’s creative liberation against a looming squeeze on software moats, white-collar work and business margins.

Oboe co-founder Nir Zicherman argues that AI learning tools need more than a chatbot’s quick answers: they need structure, multimodal lessons and the flexibility to keep learners moving toward a goal. The conversation ranges from personalized courses and the limits of LLMs to quantum mechanics, consciousness and what machines reveal about human knowledge.

Anthropic’s Claude Cowork recasts the AI chatbot as an asynchronous desktop agent, letting nontechnical users hand off browser research, document work and file tasks for longer runs. A live test of the research preview finds a promising, still-janky bridge between Claude chat and Claude Code, with its biggest questions centered on interface, permissions and trust.