|
AI Builders Digest
Sunday, November 15, 2026
|
|
The agent commerce story is moving faster than anyone expected. Three days ago we were asking who would authorize the first AI purchase. Now Replit has a Visa partnership, OpenAI has a wallet, and the question has shifted from "can agents spend money" to "who built the payment rails and who's just along for the ride."
|
|
---
|
|
01
|
Replit and Visa are building the pipes for AI commerce
|
|
|
Replit CEO Amjad Masad confirmed a formal partnership with Visa on "agentic commerce," teasing more features ahead. This follows his announcement earlier this week that Replit Agent can already buy ads on your behalf. The Visa partnership suggests Replit isn't just routing transactions through existing payment infrastructure — they're co-building something purpose-made for agents that spend money autonomously.
|
Why it matters: Yesterday we flagged that OpenAI's ChatGPT Wallet might be the infrastructure that makes this category real at scale. This complicates that picture. If Replit has Visa at the table and a working product already, OpenAI's wallet may be catching up to a race that's already started, not starting one. The question now is whether the payment layer becomes a commodity or a moat.
|
|
Source →
|
|
---
|
|
02
|
Vercel ships Workflow SDK 5 and a reusable durability layer for agents
|
|
|
Vercel CEO Guillermo Rauch announced Workflow SDK 5 and Vercel Workflows, calling it a "monumental step forward for agents." The headline improvements: faster time between agent steps, stronger encryption, and reliability proven at scale through real disaster-recovery scenarios. The most interesting piece for developers is `durable-harness`, a reusable durability log that any agent framework can build on top of, not just Vercel's own tooling.
|
Why it matters: Most agent frameworks today fail silently or inconsistently when something goes wrong mid-task. If your agent is buying ads or renewing contracts (see above), "it crashed somewhere in step 4" is not an acceptable failure mode. Vercel is making a bet that reliable, auditable agent execution is a infrastructure problem worth owning, and opening `durable-harness` to everyone is a smart way to make that standard theirs.
|
|
Source →
|
|
---
|
|
03
|
A "No AI Slop" manifesto is making the rounds, and developers are signing it
|
|
|
The Latent Space newsletter reports that a manifesto against low-quality AI-generated content, dubbed the "No AI Slop" manifesto, emerged from conference discussions at AIE CODE and was introduced by Dex in the closing keynote. It hit the front page of Hacker News and is open for signatures and edits at slopmanifesto.com. The same newsletter notes that llama.cpp and ggml both hit 1.0 this week, declaring stability after removing deprecated functions, and that LangChain closed a Series C led by CapitalG with new LangSmith features planned for November 20.
|
Why it matters: The slop manifesto is the first time you're seeing the developer community try to formalize a standard around AI output quality, not as a product feature but as a professional norm. That's a different kind of pressure on AI tools than benchmarks create. If it gets traction in engineering orgs, "does this meet slop standards" becomes a real question in code review. The llama.cpp 1.0 milestone matters separately: stability declarations from foundational open-source libraries make it easier for companies to commit to them in production without fear of breaking changes.
|
|
Source →
|
|
---
|
|
04
|
A free AI tool for cancer patients, built by one person in a weekend
|
|
|
Product builder Peter Yang shared a free AI skill called `/fuck-cancer` that lets ChatGPT or Claude work through a patient's medical records stored in Google Drive, produce a one-page summary, and research relevant clinical trials. It's open-source and forkable for other long-term conditions.
|
Why it matters: This is what the "one person with AI agents" model actually looks like in practice, not a startup pitch but a tool that would have taken a small team months to build, shipped by one person and given away. If you have someone in your life navigating a serious diagnosis, this is worth bookmarking.
|
|
Source →
|
|
---
|
|
05
|
ChinaTalk's Jensen Huang emergency pod is worth your commute
|
|
|
Yesterday we flagged this, but it deserves a second mention for anyone who skipped it: Jordan Schneider, Dylan Patel, and Jordan Nanos recorded immediately after Jensen Huang's resignation and covered the Nvidia succession question, the MATCH Act vs. RASA debate on export controls, and why no American lab has beaten DeepSeek or Kimi in open models. Dylan Patel's argument that equipment controls beat chip controls is the most concrete policy take in the conversation.
|
|
Source →
|
|
Follow builders, not influencers. A daily digest of what matters in AI.
Read online ·
Archive
|