Top pick: "Do Agents Have Network Effects?" by Andrew Chen (his own essay)

Chen is building on the horizontal-agent debate from earlier briefs, including Mike Vernal's "The Amazon Trap." He says new consumer agents such as Muse, Town, Instinct and Grokbot are forcing "many thousands of AI startups to re-evaluate their place in the market" @andrewchen. His starting argument is that agents are "tools, not networks." He compares them to email clients: when a Muse agent meets an Instinct agent, the two can talk over APIs or messaging, so nobody gains from everyone running the same agent @andrewchen.

He then lists ways an agent could build network effects anyway:

• Viral objects. Agents can create shareable microsites, boards and mini-apps that pull new users in @andrewchen.
• Shared layers. Some tasks need a common layer for identity, trust and matching. Examples include finding a passive VP of Engineering, or vetting a seller whose reputation could be gamed @andrewchen.

His conclusion is that the real question is "Where will the network effects live?" If they sit in open protocols, vertical agents can thrive. If a few horizontal agents keep identity, context and reputation to themselves, switching agents "means leaving your network behind" @andrewchen. Post

Harry Stebbings: Venky Ganesan on 20VC, and Sarah Guo's rebuttal

Stebbings posted a clip from his interview with Menlo's Venky Ganesan. In it, Ganesan says "each seed investment is an option bet," and that you size up "when there is true quantitative evidence that it is going to be an outlier." Stebbings asked how seed funds should compete when mega funds are "totally price insensitive" @HarryStebbings. Clip

Sarah Guo disagreed: "no, actually you can treat seed investments you lead as full bets, and work on them with committed belief for years… founders do that, every time. structure determines strategy" @saranormous. She is pushing back on the clip rather than endorsing it, and the exchange is worth reading for that reason.

Garry Tan: "The Post-Literate Field"

Tan quoted a post linking this Substack essay, subtitled "Marshall McLuhan, Electric Feeds, and the Rise of Modern Tribalism and Identity Politics" @RvelinEdu. His takeaway: "The ideal future media entity actually engages in both hot media (print) and cool media (pods and videos)" @garrytan. The post shown doesn't name the essay's author.

Shorter mentions

• Tim Ferriss's compilation episode on simplifying your life (his own show). It features Chip Conley, Greg McKeown, David Allen, Elizabeth Gilbert and Oliver Burkeman @tferriss. Ferriss highlighted Conley's line: "Knowledge is something you accumulate… Wisdom is something you distill, you simplify" @tferriss.
• Clem Delangue's multi-harness RL guide (Hugging Face, his own company). The method uses a capture proxy to turn coding harnesses into RL environments without modifying them @ClementDelangue. One motivating result: the same model scored 62% under Mini-SWE-Agent but 33% under Claude Code @ClementDelangue. Training in four harnesses at once improved results in all four, from 42% to 54% @ClementDelangue. Delangue says the code, data and models are all open @ClementDelangue.
• Clay Hebert's TED talk, via Matt Mullenweg. Mullenweg is in the talk's intro. He shared it and said Hebert has "been doing awesome stuff" @photomatt. He also pointed to Adam Pickering's AI-built personal site vivid-fjord.view.fast, which has a guestbook, a profile song and a butler bot @SpacefastBot. His comment: "The web is going to get weird and fun again" @photomatt.
• A YouTube video, via Elad Gil. His only caption was "Prediction of SI" @eladgil. We couldn't confirm the speaker or what is predicted, because the opening of the transcript is mostly music and applause.