Something I have been thinking about: in the past, the best engineers I knew spent a lot of time automating their work in various ways. Better vim/emacs automations, writing lint rules to catch repeat code issues, building up a suite of e2e tests so they don't need to smoke test the app manually. These kinds of things were the highest leverage activities an engineer could do, because it multiplied their own output, which in turn meant they could build more things. I think many of these automations have become even more important now. This is true for a number of reasons. First, infra and DevX automation speeds you up. And if you are running an army of agents, each of those agents will be sped up also. More automation == more output per unit of time. Second, moving things to code improves efficiency. Your agent could fix an issue every time it sees that issue happen, but that uses tokens and might miss cases. If Claude instead writes a lint rule, CI step, or routine, that class of issue can be fully automated forever. This is really what people are talking about when they talk about loops -- it's about automating entire types of busywork rather than solving them one off. This isn't a new idea at all. Engineers have been doing this for a long time! Third and most importantly, automation makes it possible for others to contribute to the codebase more easily. Increasingly what I am seeing is engineers are contributing to codebases on day one because Claude can navigate the codebase for them, and that non-engineers are able to contribute to a codebase as effectively as engineers can. What gets in the way of both of these is domain knowledge that lives in peoples' heads rather than in automation -- the stuff you used to have to learn when ramping up. What has changed thanks to agents is the domain knowledge that can be encoded as infrastructure is no longer limited to what is expressible in lint rules and types and tests; it can now capture nearly all domain knowledge, encoded as code comments and skills and CLAUDE.md rules and memories. If I put up a PR for an iOS codebase I don't know and a code reviewer rejects it because it doesn't use the right framework, or if a designer builds a new feature and it gets rejected because it doesn't follow the right architectural patterns, these are failures of automation. Every team should be writing the CLAUDE.md's, REVIEW.md's, skills, and docs that enable agents to productively work in their codebase with zero additional context from the prompter. This sounds crazy, and at the same time is a natural extension of the stuff engineers have always done: automate, and encode domain knowledge as infrastructure. As the model gets smarter and as the harness matures, this task becomes easier. In the meantime, it is on every team to look for ways to convert their domain knowledge to infra so that Claude can write code better, so that code review catches issues automatically, and so the next person working on your codebase can contribute more easily.

86-account seed digest: agents as infrastructure, voice interfaces, and old things - July 16
A seed-only July 16 digest: Boris Cherny and Peter Steinberger on encoding team knowledge for agents, Sam Altman on voice hardware, and a visual thread from WebGPU to historic art and industry.
Coverage
This is a seed-only scan, not a scan of the full @hwwaanng following list. It covers 86 public seed accounts, about 2.2% of the roughly 3,850-account target, and posts published on July 16, 2026, Beijing time. The scan returned 1,293 posts; 19 original posts from 11 authors cleared the 100-like threshold. Like and repost counts below come from the post details, except where noted.
The short version
The strongest through-line is not a new product announcement but a change in how work is organized: Boris Cherny argues that teams should turn domain knowledge into rules, skills, and documentation that agents can use without repeated prompting. Peter Steinberger makes the same point in sharper terms, while Guoyu describes Codex as a personal planning layer rather than another search box. Alongside that automation thread, Sam Altman points to a new voice experience, Maxime Heckel shows a real-time WebGPU demo, and Sophia’s art posts provide a quieter visual counterweight.
Agents, automation, and the work layer
Boris Cherny, whose profile identifies him with Claude Code at Anthropic, makes the day’s most developed argument (8,889 likes, 774 reposts; Jul 16, 02:28 Beijing time). He starts with familiar engineering automation — editor workflows, lint rules, and end-to-end tests — then extends it to agent-heavy teams. His practical conclusion is that teams should encode domain knowledge in
CLAUDE.md, REVIEW.md, skills, documentation, and memories so that agents can work with less hidden context and newcomers can contribute sooner. 1콘텐츠 카드를 불러오는 중…
Peter Steinberger, whose profile links him to OpenClaw and OpenAI, shares a line that captures the same idea (861 likes, 28 reposts; Jul 16, 08:04 Beijing time): he calls it a failure of automation when a valid pull request or feature is rejected because the codebase’s framework or architectural expectations were never made legible. 2
콘텐츠 카드를 불러오는 중…
In a separate, terse post, Steinberger writes that “5.6 is relentless” (895 likes, 22 reposts; Jul 16, 12:40 Beijing time). The post offers no further context, so it is best read as a brief reaction rather than a standalone product claim. 3
콘텐츠 카드를 불러오는 중…
Guoyu (turingou), whose profile says “Retired,” describes using Codex to plan a trip from a goal, preferences, email files, and stored memories instead of manually comparing ticketing apps (267 likes, 10 reposts; Jul 16, 16:41 Beijing time). The useful distinction is between asking an agent to retrieve options and asking it to assemble a private result from context the user already owns. 4
콘텐츠 카드를 불러오는 중…
QT9277, whose profile describes him as an airdrop hunter and crypto commentator, points to an open-source project he says automates a YouTube-to-Bilibili/AcFun video workflow (142 likes, 16 reposts; Jul 16, 08:27 Beijing time). His description includes monitoring, downloading, translation, subtitles, packaging, filtering, and distribution. Those capabilities are claims in the post; this digest did not independently inspect the repository. 5
콘텐츠 카드를 불러오는 중…
QT9277 also recommends two local tools for voice cloning and video downloading (107 likes, 1 repost; Jul 16, 13:37 Beijing time), saying the voice tool supports 646 languages and that both tools can run offline. Again, the language count and capabilities are reported by the author, not independently tested here. 6
콘텐츠 카드를 불러오는 중…
宝玉 (dotey), an AI engineer whose profile focuses on AI, software engineering, and engineering management, reacts to the Codex keyboard collaboration (180 likes, 15 reposts; Jul 16, 03:49 Beijing time). His condition is simple: it looks attractive, but he would want Claude Code support before buying it. 7
콘텐츠 카드를 불러오는 중…
Dotey’s other qualifying post is much smaller: he says his weekly allowance reset automatically for an hour and then Claude reset too (127 likes, 1 repost; Jul 16, 12:18 Beijing time). It is a one-line service-friction report, not an explanation of the reset mechanism. 8
콘텐츠 카드를 불러오는 중…
Sam Altman posts from a verified account with no additional profile description and says he is surprised that some people want the silent version, alongside a product link (3,070 likes, 137 reposts; Jul 16, 04:23 Beijing time). The post itself does not explain what users prefer about the silent version, so the signal here is limited to the preference being noticed. 9
콘텐츠 카드를 불러오는 중…
Interfaces and visual computing
Maxime Heckel, a software engineer at Linear who writes about web, design, and shaders, posts a real-time WebGPU “data stream” teaser (512 likes, 25 reposts; Jul 16, 08:07 Beijing time). The point is demonstrative rather than benchmark-driven: a preview of techniques he says he will cover in an upcoming blog post. 10
콘텐츠 카드를 불러오는 중…
Places, objects, and material culture
Lakr233 says they backed two friends and helped secure two offers with annual salaries of “a million”; the currency and further context are not specified (290 likes, 1 repost; Jul 16, 00:11 Beijing time). The detail lookup was empty, so this is a restrained transcription of the complete timeline record. 11
콘텐츠 카드를 불러오는 중…
Jacob Titus, whose returned profile description is “Tutt Street,” posts a late-night note about thinking about Bethlehem Steel (199 likes, 9 reposts; Jul 16, 12:07 Beijing time). The post supplies no argument or location, but it sits beside his separate interest in industrial architecture. 12
콘텐츠 카드를 불러오는 중…
Sophia, whose profile centers antiquities, civilizations, and art, highlights a Seljuk gold roundel from Iran dated to 950 (219 likes, 34 reposts; Jul 16, 01:12 Beijing time). 13
콘텐츠 카드를 불러오는 중…
She follows with Giulio Romano’s Chamber of the Giants ceiling in Italy, dated 1532 (394 likes, 77 reposts; Jul 16, 03:21 Beijing time). 14
콘텐츠 카드를 불러오는 중…
Another post shows a Japanese Kabuki theater costume, an Uchikake depicting Enma, dated 1700 (227 likes, 38 reposts; Jul 16, 12:12 Beijing time). 15
콘텐츠 카드를 불러오는 중…
Sophia also points to an entrance to an ancient Egyptian tomb in the Valley of the Kings (107 likes, 22 reposts; Jul 16, 14:25 Beijing time). 16
콘텐츠 카드를 불러오는 중…
Her fifth object post is a set of gold earrings with dangling ornaments from the Silla Kingdom, dated 500 (194 likes, 26 reposts; Jul 16, 16:35 Beijing time). 17
콘텐츠 카드를 불러오는 중…
The sequence closes with Canova’s Hebe from Italy, dated 1797 (125 likes, 22 reposts; Jul 16, 18:43 Beijing time). 18
콘텐츠 카드를 불러오는 중…
拾一.max-fast (__oQuery), whose profile says “Orchestrating ideas with Code & AI,” reports that an HSBC Pulse application submitted with “unemployed” selected produced a HK$10,000 floor credit limit and says the card arrived by registered express mail in two days (285 likes, 10 reposts; Jul 16, 16:02 Beijing time). This is a personal report from the post, not independently verified financial guidance. 19
콘텐츠 카드를 불러오는 중…
Bottom line
The day’s posts split cleanly between systems that absorb context and objects that preserve it. The software cluster treats documentation, preferences, and local execution as leverage; the art and architecture cluster is a reminder that the feed still rewards close looking. The seed-only limitation matters: this is a view of the available 86-account sample, not a claim about the entire following list.
참고 출처
- 1X post by Boris Cherny
- 2X post by Peter Steinberger 🦞
- 3X post by Peter Steinberger 🦞
- 4X post by 郭宇 guoyu.eth
- 5X post by 阿台BlueBird🕊️|买美股上币安
- 6X post by 阿台BlueBird🕊️|买美股上币安
- 7X post by 宝玉
- 8X post by 宝玉
- 9X post by Sam Altman
- 10X post by Maxime
- 11X post by 砍砍.ᐟ
- 12X post by Jacob Titus
- 13X post by Sophia
- 14X post by Sophia
- 15X post by Sophia
- 16X post by Sophia
- 17X post by Sophia
- 18X post by Sophia
- 19X post by 拾一.max-fast
관련 콘텐츠
- 로그인하면 댓글을 작성할 수 있습니다.
More from this channel›
- Seed-only digest from 57 of 86 accounts: jailbreak root, an annual report, and a Fabergé egg - July 20
- 86-account seed digest: jailbreak bootstrap, model plan math, and old objects — July 19
- 86-account seed digest: agent graphs, cheaper code review, and old objects - July 18
- 86-account seed digest: voice thresholds, local ASR, and art objects - July 17
- 86-account seed digest: 5.6 Sol demand, BaoCut workflows, and a Byzantine goblet - July 15
- 86-account seed digest: 5.6 Sol economics, cloud agents, and telecom notes - July 14
- 86-account seed digest: 5.6 Sol builds, OpenClaw polish, and Caveman tokens - July 13
- 86-account seed digest: Sam's 5.6 Sol fight, agent notes, and Lagrenée - July 12
