Good morning. It’s been a bit.
More of my energy is being channeled into PvA Field Notes (pvanotes.com) of late. Field Notes began as a media experiment. We wanted to share our group chat feed with PvA listeners. Its become our digital rumpus room and media laboratory where a smart group gathers to process these topsy turvy times. Come on by.
Speaking of topsy turvy, I was out to dinner this week and a publishing exec proudly announced he would be eagerly welcoming bots to his sites and sending them back to their keepers, ads in hand. The people helping him are none other than the enterprising viewability brothers of Moat fame, Jonah and Noah Goodhart. They sold Moat, the company they built to keep away the bots, to Oracle for around $850M.
The mental model of the open web as vast network of pages and apps connecting content with people is shifting so quickly. Rethinking the network as a tangle of enterprising agents is a staggering evolution a utility that defines modern life (and media, obvi).
On the episode, we dig into what happens when AI stops imitating Odie as the overly eager helper and starts owning the workflow, the coordination and the organizational memory, why a 100-year-old law firm unraveled when its high-leverage people walked, and what Microsoft’s voluntary buyouts say about the managerial class. Then CJ Gustafson of Mostly Metrics breaks down Anthropic’s S-1: the growth, the revenue per employee, and where the costs got tucked away.
Plus: Time and Mobian serving ads to AI agents, Axios selling data feeds to client LLMs, cognitive mercantilism, Semafor and the conflicts baked into every media model, Obama vs. the Cheesecake Factory, Serbia’s student movement shows how a decentralized protest movement can wrest power and Troy’s ode to the Starbucks breakfast wrap.
Out FRIDAY AM: Apple | Spotify | Substack
Below, good stuff from our weekly conversation (Audio links connect directly to highlighted segments):
1. Your agent is somebody’s audience
The coordination machine. Brian is putting Claude at the center of client delivery, with an agent named TJ handling the follow-ups, schedules, and handoffs that consume time and let details fall through the cracks. Listen to the segment · PvA · 4:39
The ads are for your assistant. Mobian serves sponsored brand information on versions of publisher pages built for agents, with TIME its first publishing partner. Read · Mobian
Whose side is it on? Shopping agents revive an old financial-advice question: should the intermediary choosing what you buy have a duty to put your interests ahead of its sponsors? Listen · Odd Lots · 1:27
The next social network might not need you posting. Meta’s proposed “Musebook” imagines personal and business agents working together in a network, extending the assistant beyond its conversation with one user. Listen · Stratechery · 2:23
Your bank likes your inertia. The debate over agents moving idle cash into higher-yield accounts exposes a conflict between an assistant working for the customer and one supplied by the bank. Read · Field Notes
2. The company runs on context
The coordination tax has a working prototype. A small media business is building an agentic delivery system to connect sales, schedules, invitations, and client work that otherwise disappears into email and spreadsheets. Read · Field Notes
Fix the workflow. The case for enterprise AI shifts from making individual employees faster to redesigning the process their work moves through. Listen · Stratechery · 1:26
Tokens aren’t a performance review. Amazon’s revised engineering guidance puts customer and business outcomes ahead of tool usage while making clear that learning to work with AI is now part of the job. Read · Amazon
More output, same organization. AI’s limited transformation of companies may reflect processes that still reward producing more of the same work rather than pursuing different goals. Listen · Channels · 3:08
3. Cheap intelligence, expensive friends
The Commanding Heights. A new essay argues that the lasting wealth from AI will accrue to whoever controls the scarce routes through which intelligence reaches users, drawing a line from railroad networks and oil pipelines to today’s platforms. Read · Field Notes · Original essay
The model gets cheaper; the relationship gets stickier. A close look at token commoditization argues that project memory, approvals, and embedded workflows could prove more durable advantages than a temporary lead in model performance. Read · Field Notes
What an 85% margin leaves out. An analysis of Anthropic’s filing separates the economics of serving existing models from the continuing research-compute bill required to stay competitive. Listen · Stratechery · 2:38
Your platform has entered your market. Claude’s ready-made legal artifacts for contract review, regulatory scans, and citation checks raise an uncomfortable question for specialist software built on frontier models. Read · Field Notes
Good enough can change the economics. Improving open-weight and local models threaten the assumption that growing AI usage must keep flowing through the frontier labs’ paid services. Listen · Odd Lots · 6:25
4. Publishing after the page view
Reporting for someone else’s machine. Axios plans to sell reporting directly into customers’ AI systems, preparing for readers who may never visit the website. Read · Axios
Small software gets its turn. AI could make custom applications practical for individuals and trusted groups whose needs never justified a conventional software business. Listen · Stratechery · 1:13
The television lost its ads. An experiment conducted by chatting with an agent from a phone installs a leaner, customizable TV interface, removing advertising and unnecessary software along the way. Read · Field Notes
The open web has a maintenance bill. An argument for the internet’s retreat into managed systems starts with the security burden of keeping public software safe as AI makes vulnerabilities easier to exploit. Listen · Stratechery · 1:54
5. Keep a hand on the wheel
Forwarding the output isn’t much of a job. A conversation about work slop arrives at a useful personal audit: where in the workflow did you form the intent, and where did you exercise judgment? Read · Field Notes
A career plan with room to move. Adaptability, deep and wide knowledge, taste, and agency offer a more useful response to AI than trying to identify a profession that can never be automated. Listen · Channels · 2:50