Seed-only August 22 X digest: agent workflows, Fortune fragments, and 10 posts from 5 authors

This edition covers August 22, 2026, 00:00–24:00 Beijing time. The scan checked 86 configured public seed accounts, returned 654 posts, found 105 posts in the window, excluded 17 retweets, and kept 10 original posts from 5 authors with at least 100 likes. The full @hwwaanng following list remains unavailable, so this is a roughly 2.2% sample of the intended network.

At 00:50, 宝玉 (@dotey) recommends using Fable's high mode mainly for orchestration and acceptance, while delegating implementation work to Opus or Sonnet subagents. The attached prompt assigns the main agent requirements clarification, task breakdown, delegation, and result checking, while subagents handle code reading, writing, testing, and batch edits. The post had 375 likes and 41 reposts. 1

At 06:46, 宝玉 describes an ELI5 Skill: a prompt that asks for an HTML explanation aimed at someone with no background, using large images and few words. He says the Skill is essentially that one prompt and may be unnecessary for people who would use it rarely. The post led the day's sample with 888 likes and 144 reposts. 2

At 23:01, 宝玉 relays an argument about the "last 5%" of work. The post compares AI coding tools with a dishwasher: standardized work becomes cheaper, while edge cases, integration, and business understanding still require people. It also argues that a larger software industry could create more of those edge cases. The post had 252 likes and 36 reposts. 3

At 17:35, 空谷 Arvin Xu (@arvin17x) compares Pi's planned database-style direction with the distributed, serverless runtime he says LobeHub built for cloud execution. He says Pi assumes one writer per session, while his own setup must handle multiple workers competing for one operation. The post claims support for roughly 1,000 concurrent agents per user, runs of more than 3,600 steps, and agent scheduling costs 10 times lower than the industry; those figures are claims made in the post. It had 147 likes and 6 reposts. 4

At 04:26, Peter Steinberger (@steipete) says he spoke at the Agentic AI Summit in Berkeley and that his talk was titled "No Doors for Agents." The post gives the event and talk title, but no argument or technical detail. It had 269 likes and 17 reposts. 5

At 07:36, Peter Steinberger posts, "dropping new skill brb." The post names no Skill and gives no purpose, so it is a short release signal rather than a description. It had 307 likes and 10 reposts. 6

At 00:53, Jacob Titus (@jacob__titus) posts the line "Fortune made a world of American business. Workers." and a link. The tweet supplies no further explanation; its value for a morning scan is the pointed one-line reference itself. It had 354 likes and 11 reposts. 7

At 06:11, Jacob follows with "Fortune made a world of American business. Machines." and a second link. The post again leaves the surrounding argument to the linked material. It had 132 likes and 7 reposts. 8

At 17:31, 郭宇 guoyu.eth (@turingou) writes, "Sometimes you have to admire the imagination of these Silicon Valley startups..." The post names no startup, product, or argument, so there is little to assess beyond the reaction itself. It had 269 likes and 20 reposts. 9

At 17:39, 郭宇 guoyu.eth writes, "Building software from scratch is easier than ever, yet more meaningless than ever." The post offers no example or explanation, leaving the sentence as a standalone judgment. It had 306 likes and 12 reposts. 10