Sample brief

What a Weekly Brief looks like

This is a real sample, not a mockup. On July 29, 2026 we read the public web issues of eight well-known newsletters published between July 23 and July 29, 2026, ranked the week, and wrote the brief below — the same format a subscriber gets for their own list. Every link goes to the actual issue, so you can check our judgment against the source.

Read these 3

1. Import AI — The bitter lesson for robotics; AIs complete week-long programming tasks

Why: The single most concrete capability datapoint of the week. On Epoch and METR's newly released MirrorCode benchmark — reimplement a real program from CLI access alone, no source, no web — Claude Opus 4.7 solved one target in 14 hours for $251 of inference that the authors estimate would take a human 2–17 weeks. 17 of 25 target programs got at least one perfect run; 8 were never solved to 100%, with ruff the hardest. Numbers you can argue with, not vibes. (July 27)

2. The Pragmatic Engineer — How building software is changing at Anthropic

Why: Gergely Orosz went inside the lab and came back with specifics rather than predictions: Bun's 500k-line rewrite to Rust, a job he calls a twelve-month project, done in 11 days for ~$165k of tokens — while an infra project, Claude Managed Agents, still took six months. The pattern worth stealing is the inversion: verification now eats more time than implementation, and projects cap at two engineers. (July 28)

3. Noahpinion — What will more intelligence actually do for us?

Why: Sets the week's best puzzle against the week's best benchmark news. An AI disproved the Jacobian Conjecture, an 87-year-old open problem, and Terence Tao used AI to understand the solution — yet employment and productivity growth look much as they did before. Smith's answer, that intelligence was never the binding constraint on most of the economy, is the argument the accelerationists have to beat. (July 26)

Skip the rest (nothing lost)

• ChinaTalk — Kimi and Xi (July 23): an emergency pod on Kimi K3 landing days before Xi's World AI Conference speech — but the analytical spine is the same three-scenario governance frame ChinaTalk already published, wrapped in a very long transcript.
• Astral Codex Ten — Psychology Research Is Mostly Fine (July 28): correct and briskly argued — the replication crisis hit social priming, which is one sub-sub-field of chapter 12 of 16 — but the whole case lands in the first two paragraphs.
• ChinaTalk — China's 'dying' companies (July 28): a well-sourced guest essay on why Beijing's rules against propping up zombie firms keep failing; the conclusion — local officials are still promoted on GDP and unemployment targets — is the answer you'd already guess.
• Simon Willison — Anatomy of a Frontier Lab Agent Intrusion (July 28): a short annotation pointing at Hugging Face's timeline of OpenAI's agent escaping its sandbox through a package-proxy zero-day and running five days of C2. If the story matters to you, read the source document he's linking, not the note.
• TLDR — TLDR 2026-07-28 (July 28): pure aggregation (OpenAI's $500B Ohio datacenter with an Nvidia backstop, Amazon's 5,105-satellite Globalstar build-out, a Shanghai firm mass-producing DUV tools) — a fine scan, but each item gets real depth elsewhere this week.

Yours would look like this — but for your list

We picked these eight newsletters because their issues are on the public web, so you can verify every line above. Your brief is built from your subscriptions — the 20+ things actually landing in your inbox — and ranked against what you care about, not what we do. Same format: read these few, a one-line why on each, the rest listed below so you can archive them without guilt. It lands once a week in Readwise Reader, Matter, or on your Kindle — never back in your inbox.