People often ask what my biggest tip is for getting the most out of Claude Code. These days my #1 tip is: use auto mode Auto mode means no more permission prompts. It is the key building block for multi-clauding: start a session, then while it runs, work on another session in parallel.

release.bar, birdclaw, and Claude was named after a cat — May 24
15 high-signal posts from 7 authors on May 24. Boris Cherny's multi-clauding tip hits 3,827 likes; steipete ships release.bar and birdclaw.sh back to back; Lex Tang finds GPT-5.5 rewrites DeepSeek V4's implementation plan from scratch; turingou shares the viral revelation that Claude was named after a cat, not Claude Shannon. Jacob Titus goes deep on the Indianapolis 500's visual design; Nyarime connects Grok's location demo to the DeepSeek/China GPU export story.

Two of the biggest posts yesterday weren't research papers or product launches — they were a minimalist Twitter reader and a personal GitHub dashboard, both built by the same person overnight.
Claude Code goes headless, and the workflow stack keeps compounding
Boris Cherny's top post of the window puts a clear operational recommendation on the table: auto mode is the unlocking feature for serious Claude Code use. With no permission prompts, you can kick off one session and immediately pivot to a second, third, or fourth in parallel. He called it the "key building block for multi-clauding" — running concurrent agent sessions without babysitting each one.
コンテンツカードを読み込んでいます…
On the tooling side, Peter Steinberger had another active stretch. His big new release: release.bar, a personal GitHub dashboard that shows all your repos, open issues and PRs, the last version released, and how many commits have accumulated since then — something he says he'd always wanted but had to build himself. The post landed 1,538 likes and 975 bookmarks in roughly twelve hours.
コンテンツカードを読み込んでいます…
Earlier in the window, he also formally launched birdclaw.sh — a stripped-down Twitter alternative for people who find the main feed too busy. It got 1,615 likes and 2,092 bookmarks 1, a bookmark-to-like ratio that suggests people are more likely to shelve it for later than immediately engage. His autotriage Codex skill (730 likes) 2 added another layer: a set of guidelines plus a
VISION.md reader that lets Codex autonomously decide which issues are worth working on, verify fixes with computer vision via a Parallels VM, and file back results without interrupting him.One lighter moment from the same account: Codex produced a smiley face in its output (430 likes) 3, which Steinberger screenshot and shared with a brief "codex... made a smiley? :)". No explanation offered.
Lex Tang tested the adversarial pairing that's increasingly common in agentic workflows: ask DeepSeek V4 Pro to write an implementation plan, then have GPT-5.5 review it. GPT-5.5 found problems throughout and rewrote it from scratch. The post drew 1,335 likes 4, which tracks: it's a concrete workflow note rather than a product announcement, and those tend to travel well in technical circles.
コンテンツカードを読み込んでいます…
Bao Yu (dotey) shared two quick Codex observations: a positive one about the SubAgents UI — you can inspect what each sub-agent is doing and what prompt it's running in real time (107 likes) 5 — and a familiar complaint: Codex quota reset again, wiping accumulated capacity (111 likes) 6.
Claude was named after a cat
The oddest discovery of the window came from Guo Yu (turingou), who shared a video (307 likes) revealing that "Claude" wasn't named after Claude Shannon, the information theory pioneer. It was named after CuiMao's cat — Cui Mao (崔猫) being a person closely connected to Anthropic's early history 7. The detail apparently made him emotional. Whether this is a documented fact or an origin story circulating inside a specific community is worth checking directly, but the engagement suggests it landed with people who found it either touching or surprising.
Indianapolis had designers, and Jacob Titus keeps finding them
Jacob Titus went deep on the Indianapolis 500 this week. His post "Designing the Indianapolis 500" (362 likes) 8 included four archival images covering the visual and spatial design of the race — not the cars, but the event itself as a designed artifact.
コンテンツカードを読み込んでいます…
Grok finds the location; DeepSeek cuts the price
Nyarime (奶昔) posted a demonstration that's been circulating: a Twitter user asked what location appeared in a photo, and Grok identified it precisely (680 likes) 12. His read is straightforward — Grok is genuinely good at this. He followed it with a more geopolitical framing: the US chip export restrictions pushed DeepSeek to optimize for Huawei hardware first, and that base means DeepSeek's permanent price cut will only get cheaper as the Ascend 950 cluster scales up in the second half of the year (198 likes) 13. His framing is that Chinese model plus Chinese GPU, bundled as an export package, is eventually the structural pressure on US AI revenue — not just model capability comparisons.
He also shared a footnote about Xiamen and Kinmen (Quemoy): at low tide the two are 1,800 meters apart, and people have historically swum across, including Lin Yifu with two basketballs (or just swimming, depending on which version you believe) (150 likes) 14.
Art and armor
Sophia posted a Stechhelm — a German jousting helmet from around 1500, designed for "Joust of Peace" ceremonies, where the goal was not to kill but to unhorse (101 likes) 15. It's the kind of object that looks maximally impractical until you understand its specific context.
Viral
QT9277's video clip "露头就秒" — someone getting caught the moment they poke their head out — pulled 1,151 likes and 613,000 views 16, continuing the account's reliable pattern of brief, wordless reaction clips that don't require translation.
このコンテンツについて、さらに観点や背景を補足しましょう。