Sunday, August 16, 2026

Good morning, NOLA. Sunday’s brief is less about another model launch and more about the craft of using the ones we have: a sharp essay argues that working with AI is becoming a leadership skill, while Simon Willison’s CORS Chat is a tiny, useful way to test a local model. There’s also a grounded look at AI drug discovery, new detail on Claude’s watermarks, and an unexpectedly delightful game that makes you roleplay as the chatbot.

Using AI Like a Teammate

Working with AI feels more like leadership than coding

The useful frame here is not “ask AI for code,” but set direction, supply context, review the work, and course-correct. That should resonate whether you are building software, a sales workflow, or an internal tool: the bottleneck is increasingly good delegation. The Hacker News discussion has a lively companion conversation.
Hacker News

Why AI’s huge context window is not the same as mathematical insight

A thoughtful corrective to the “AI is smarter than everyone” shorthand. The essay argues that models can hold and combine far more working material than a person, but that is different from choosing the right abstraction or deciding what matters. HN’s discussion is worth a scan for builders thinking about where human judgment still belongs.
Hacker News

A model raised only on fifth-grade material

This experiment asks a charmingly direct question: what kind of language model emerges when its training material stops at fifth-grade level? It is a good weekend read about how much of apparent intelligence comes from knowledge, language exposure, and the examples we give a model. Discussion on HN.
Hacker News

AI drug discovery: progress, limits, and what comes next

A sober, accessible status check on one of AI’s most promising real-world applications. The key takeaway for non-biologists: AI can make parts of the discovery process faster, but proving that a drug works remains the hard, physical-world part. HN discussion.
Hacker News

Tools, Testing, and Agent Workflows

CORS Chat: a quick local-model testing tool

Simon Willison built a small browser tool for testing a locally running model from another web app. If you have been trying the Qwen release we featured yesterday in LM Studio, this is a practical little companion: it helps verify that your local API can accept browser requests without building a whole demo first.
Simon Willison

Yadda 3.0 brings AI into behavior-driven testing

Yadda’s update aims to make plain-language test scenarios more useful in an agent-assisted development workflow. The practical idea is sound: give an agent clear examples of what “done” looks like, then keep those examples as tests so the next change does not quietly break them. HN discussion.
Hacker News

Flue 2 borrows React’s hooks for AI agents

Latent Space talks with Astro creator Fred Schott about Flue 2, a framework for giving agents repeatable workflows and checkpoints. It is more builder-facing than plug-and-play, but the larger lesson is approachable: agents get more dependable when their work is structured instead of being one giant prompt.
Latent Space

Computer History records actions instead of endless screenshots

The latest AI Daily Brief episode highlights Computer History, which creates a record of computer interaction from events rather than a constant stream of screenshots. That could make it easier for AI assistants to understand what happened in a workflow while keeping the record lighter and more navigable.
AI Daily Brief

Product Moves and Creative Detours

Anthropic explains how Claude’s new watermarks will work

Yesterday we linked a plain-English guide to AI text watermarking; this is the concrete follow-up on Claude’s implementation. The important builder question is whether markers survive common edits and how they apply to code, not the underlying detection mechanics. Pair it with yesterday’s visual guide for the approachable version.
TechCrunch AI

SpaceX officially closes its Cursor acquisition

The deal is now official: Cursor joins SpaceX. For everyday users, the immediate product implications are still unclear, but it is a notable signal that AI-assisted coding has become strategic infrastructure for companies building ambitious hardware and software together.
TechCrunch AI

Roleplay as a chatbot, and discover how strange the job is

Your AI Slop Bores Me flips the usual interface: one person writes a request and the other has to answer as the machine. It is a funny, low-stakes creative experiment—and a surprisingly good prompt-writing lesson once you experience vague requests from the other side.
The Verge AI

Worth a Listen

The AI Deputization Audit: deciding what work to hand to AI

If you missed it in yesterday’s quick scan, this AI Daily Brief episode deserves the promotion. It offers a useful audit for deciding which tasks should be delegated to AI, which should stay human-led, and where a review step is non-negotiable.
AI Daily Brief

Also

Today’s Sources