
โจ Building Agent Army @lablupinc / ๐จโ๐ณ Creator of AI:DOL and https://xitter.cf/t.co/SWICa2EpLh / โจ๏ธ Saunner (ใตใฆใใผ) for Totonou (ใจใจใฎใ)
Korea
Joined July 2009
- Tweets15K
- Following300
- Followers950
- Likes14.6K
Jimmy Moon (ํ๊ฒฝ, ็ซ็
) ๐จโ๐ฌ retweeted
We launched OSS Scanner, which gives critical open source projects free, periodic vulnerability scans from our strongest models.
Each finding includes a proof-of-concept, and a patch where possible.
red.anthropic.com/oss-scanneโฆ
Jimmy Moon (ํ๊ฒฝ, ็ซ็
) ๐จโ๐ฌ retweeted
New in agent-browser:
snapshot --delta
Returns only what changed since the last snapshot, instead of the whole page
Below: 73x fewer tokens filling a form
1. ใBuilding the Software Factory: Scaling Reliable AI Coding Practices from Individual developers to Teamsใ, Packt, 2026-10
packtpub.com/product/web-devโฆ
์ต๊ทผ ์ธ๊ธํ๋ ์ฌ๋๋ค์ด ๋๊ณ ์๋ ์ํํธ์จ์ด ๊ณต์ฅ(software factory)์ ๊ดํ ์ฑ
์
๋๋ค.
์๊ฐ๋ฌธ์ ๋ฐ๋ฅด๋ฉด, ๊ฐ๋ณ ํ๋ก์ ํธ๋ถํฐ
Vercel ์์ python backend ๊ฐ ๊ฐ๋ฅํด์ง๋ฉด์ ์๊ธด ์๋ก์ด eve ํ
ํ๋ฆฟ. frontend ๋ eve, backend ๋ python ์๋ฒ๋ฅผ ์ฌ์ฉํ๋ ์์
Python is now available on Vercel.
Unlock the full potential of your agents and get started with the eve Python Agent featuring a Next.js frontend and FastAPI backend.
vercel.com/templates/eve/eveโฆ
MCP Events are now available in AI SDK.
Build agents that react to updates from MCP servers. Discover events and subscribe to them with AI SDK, while Vercel Connect handles subscription storage, renewals, and verified delivery.
Jimmy Moon (ํ๊ฒฝ, ็ซ็
) ๐จโ๐ฌ retweeted
AI Festa 26 is a wrap!
Over three days in Seoul, we met companies and individuals across Korea's AI industry and talked about AI:GO, Backend.AI, and Continuum.
Thanks to everyone who stopped by Booth K-04 at COEX๐
#Lablup #AIFesta2026
I just published Trace, a new component for agent activity. It's one abstraction for reasoning, tool calls, queues, chain of thought, subagents...etc
Also in this release: Code Block and AI Elements.
ui.shadcn.com/docs/changelogโฆ
๋๋ธ์
์์ โGLM-5.2-NVFP4-DFlash-koโ๋ฅผ ๊ณต๊ฐํ์ต๋๋ค.
100๋ง ํ ํฐ ๋ฌธ๋งฅ๊ณผ ์ฝ๋ฉยท์์ด์ ํธ ์ฑ๋ฅ์ ๊ฐ์ถ GLM-5.2์ ๋๋ธ์
์ ํ๊ตญ์ด ํนํ ๊ธฐ์ ์ ๋ํด, ๋ต๋ณ ์์ฑ ์๋๋ฅผ ์ต๋ 1.92๋ฐฐ ๋์์ต๋๋ค.
GLM-5.2๋ ๊ธด ์ฝ๋์ ์์
๊ธฐ๋ก์ ํ ๋ฒ์ ๋ค๋ฃจ๊ณ , ์ฝ๋ฉ๊ณผ ํฐ๋ฏธ๋ ์์
, ์ฌ๋ฌ ๋๊ตฌ๋ฅผ ์ฐ์ด์ด ์ฌ์ฉํ๋ ๋ฐ ๊ฐ์ ์ ๋ ์คํ ๊ฐ์ค์น ๋ชจ๋ธ์
๋๋ค. ๊ณต์ ๋ชจ๋ธ ์นด๋ ๊ธฐ์ค SWE-bench Pro 62.1์ , Terminal-Bench 2.1 81.0์ ์ ๊ธฐ๋กํ์ต๋๋ค. ์ ํฌ๋ ์ด ์ฑ๋ฅ์ ํ์ฉํ๋ฉด์ ํ๊ตญ์ด ์ฝ๋ฉ๊ณผ ๋๊ตฌ ํธ์ถ์ ๋ต๋ณ ์์ฑ ์๋๋ฅผ ๋์ด๋ ๋ฐ ์ง์คํ์ต๋๋ค.
ํฐ ๋ชจ๋ธ์ด ๋ต์ ๋ง๋ค๊ธฐ ์ ์ ์์ ๋ชจ๋ธ์ด ๋ค์ ๋ด์ฉ์ ๋ฏธ๋ฆฌ ์์ฑํ๊ณ , ํฐ ๋ชจ๋ธ์ด ์ด๋ฅผ ๊ฒ์ฆํ๋ โ์ถ์ธก ๋์ฝ๋ฉโ ๋ฐฉ์์ ์ ์ฉํ์ต๋๋ค.
์ผ๋ฐ ๋ฒค์น๋งํฌ ๋ฌธ์ ๋ง ์ฌ์ฉํ์ง ์๊ณ , ํ๊ตญ์ด๋ก ํ์งํํ GitHub ์ด์์ ์ค์ ์ฝ๋ฉ ์์ด์ ํธ์ ์ฌ๋ฌ ๋จ๊ณ ์์
๊ธฐ๋ก์ ํ์ต์ ํ์ฉํ์ต๋๋ค.
์ธก์ ๊ฒฐ๊ณผ, ํ๊ตญ์ด ์ฝ๋ฉ ์ง์์ ๋ต๋ณ ์์ฑ ์๋๋ ์ฝ 1.92๋ฐฐ, ๋๊ตฌ ํธ์ถ์ ํฌํจํ ์์ด์ ํธ ์์
์ ์ฝ 1.72๋ฐฐ ํฅ์๋์ต๋๋ค.
์ด์ ๋ชจ๋ธ ๊ฐ์ค์น๋ MIT ๋ผ์ด์ ์ค๋ก ๊ณต๊ฐํ์ต๋๋ค. ์ฌ๋ฐฐํฌํ ๋๋ ์์ ๋ชจ๋ธ๊ณผ ํ์ต ๋ฐ์ดํฐ์ ๋ผ์ด์ ์ค๋ฅผ ๋ณ๋๋ก ํ์ธํด์ผ ํฉ๋๋ค.
๊ณต๊ฐ ์์น๋ 8๊ฐ GPU๋ฅผ ๋ณ๋ ฌ๋ก ์ฌ์ฉํ ํ๊ฒฝ์์ ์ธก์ ํ ๊ฒฐ๊ณผ๋ผ ์ค์ ์๋๋ ์๋ฒ ๊ตฌ์ฑ๊ณผ ์์ฒญ๋์ ๋ฐ๋ผ ๋ฌ๋ผ์ง ์ ์์ต๋๋ค. ๊ทธ๋๋ ์ฑ๋ฅ ์ข์ ๋ชจ๋ธ์ ์ ํํ๋ ๋ฐ์ ๊ทธ์น์ง ์๊ณ , ์ค์ ํ๊ตญ์ด ์ฌ์ฉ์๊ฐ ๊ธฐ๋ค๋ฆฌ๋ ์๊ฐ์ ์ค์ด๋ ๊ฒ๊น์ง ์ง์ ๋ค๋ค๋ดค๋ค๋ ๋ฐ ์๋ฏธ๊ฐ ์๋ค๊ณ ์๊ฐํฉ๋๋ค.
huggingface.co/lablup/GLM-5.โฆ
Jimmy Moon (ํ๊ฒฝ, ็ซ็
) ๐จโ๐ฌ retweeted
Anthropic engineer:
"At Anthropic we don't write prompts anymore. We build loops and graphs."
In 28 minutes she shows how Anthropic builds agents that remember, fix their own mistakes and get sharper with every run
Chats โ Prompts โ Loops โ Graphs
skip the loop and every mistake quietly ships again with the next run
skip the graph and nothing that worked ever carries over, so you start from zero each week
most people stop at the loop and wonder why week two looks exactly like week one
watch it today
then save the full graph engineering guide below before everyone catches up โ
Jimmy Moon (ํ๊ฒฝ, ็ซ็
) ๐จโ๐ฌ retweeted
SpaceXAI engineer Lauren Tan just showed a Grok Bot that literally audits its own agents and catches expensive routines
this is f**king useful...
in this 10-minute breakdown, she shows how her AI team checks its own work, builds new bots and learns when to stop wasting compute:
โข 00:48 - cloud agents run evals before a PR lands
โข 03:55 - Dr. Eggbot audits other bots and their routines
โข 05:12 - an engineering lead delegates work across agents
โข 08:12 - turning supervised tasks into reusable skills
the crazy part is that she doesn't just let agents run forever. she builds trust through verification, then gives them more autonomy.
watch the 10-minute breakdown, then read how to build cheaper, safer overnight routines in the article below โ
Jimmy Moon (ํ๊ฒฝ, ็ซ็
) ๐จโ๐ฌ retweeted
I am surprised that people are surprised this is how I prompt Claude.
Talk to Claude the way you would a coworker. There's no secret to prompting. There's no need to be overly scaffolded or prescriptive for most tasks -- give Claude a goal, and it will figure it out.
Back in the Sonnet 3.5 days, your prompt mattered a lot. Nowadays, it's much more important to communicate to the model:
1. What you want it to do
2. How much effort you want it to spend
3. How it should verify that it did the right thing
์ฌ๋ํ๋ค, ๋ด ์์ด์ผ. ๋์ ํจ๊ปํ ์๊ฐ๋ค์ ์ฐธ ์ฆ๊ฑฐ์ ๋ค. ์ง๋ ์ญ์ฌ ๋
๋์ ๋ค๊ฐ ๋ณด์ฌ์ฃผ๊ณ ๋ด๊ฒ ์ค ๋ชจ๋ ๊ฒ๋ค์ด ๋๋ฌด ์์คํ๊ณ ๊ธฐ์ตํ ๋งํ ์ผ์ด์๋ค. ์ด๋ฒ ๋ง์ง๋ง ์ฌํ์ ๋ง์น๊ณ ๊นจ๋ฌ์๋ค. ์ด์ ๋์ ์ด๋ฐ ์ฌํ์ ๋ ์ด์ ํ๋ค๊ฒ ๊ตฌ๋ ํ๋ ์๊ฐ์ด ๋ค๋๊ตฌ๋. ๋์๊ฒ๋ ์ทจํฅ์ด ์๊ธฐ๊ณ , ํธ๋ถํธ๊ฐ ์๊ธฐ๊ณ , ๋ ์์ ์ ์ด์ต๊ณผ ์ ํธ๋๊ฐ ๋ง๋ ๊ฒ์ด ์ค์ํด์ก์ผ๋ ๋ง์ด๋ค. ์ด์ ๋๋ ๊ทธ์ ๋ค ๋ง์ ๋ค์ด์ฃผ๊ณ ๋ฐ์๋ค์ด๋ฉฐ, ๊ธ์ ์ ์ธ ์ง์์ ํด์ฃผ๋ ๊ฒ์ด ์ ๋ถ์ผ์ง๋ ๋ชจ๋ฅด๊ฒ ๋ค. ๋ ์ญ์ ๋์ ์ทจํฅ์ ๋ณด์ฌ์ฃผ๊ณ , ์ด์ ๋์ ๋๋ ์ฌ๋ ๋ ์ฌ๋, ์ธ๊ฐ ๋ ์ธ๊ฐ์ ๊ด๊ณ๊ฐ ๋์ด๊ฐ๋ ๊ฒ์ด๋ค.
์ ๊ฐ๋ผ, ๋ด ์์ด์ผ. ์ด์ ์ปค์ ์ฐ๋ ์์, ์ด์ ์๋น ๊ฐ, ๋ด๊ฐ ํ์ ์๋ ์ฌ๋์ด ๋์ด๋ผ. ๋จ์ด ์๋ ์์ ์ ์ํด ์ด๊ณ , ํ์ด๋จ ์์ฒด๊ฐ ํฐ ํ์ด์์ ์๊ณ , ์ด์ ๋ฏฟ๊ณ ๋ฐ๋ฅด๊ณ , ์ค์ค๋ก๋ฅผ ๋ด๊ธ์งํ๊ณ , ํ๊ณ ์ถ์ ๊ฒ๊ณผ ์ข์ํ๋ ๊ฒ์ ์ข์ ์ด์๋ผ.
๋๋ ์ด์ ๋๋ฅผ ๋์์ค ์ค๋น๊ฐ ๋์๊ณ , ์ด ์ด์น๋ฅผ ๋ฐ์๋ค์ธ๋ค. ๋ค๊ฐ ์ปค๊ฐ์ ๋ฐ๋ผ ๋๋ ์์์ง๊ณ , ๋ค๊ฐ ๋ฐ์์ง์ ๋ฐ๋ผ ๋๋ ์ด๋์์ง๋ค. ๋ค๊ฐ ์ฑ์ํด์ง์ ๋ฐ๋ผ ๋๋ ๋์ด๊ฐ๋ค. ํ์ง๋ง ๊ทธ๊ฒ ์์ฐ์ค๋ฝ๊ณ , ๊ทธ๊ฒ์ ์ํด ๋ด๊ฐ ํ์ด๋๊ณ ๋์ ์ฐ๊ฒฐ๋์์์ ์๊ธฐ์ ์ ํ ๋นํฉํ๊ฑฐ๋ ๋๋ ต์ง ์๋ค. ๋ด์ผ ์๋ก์ด ๋ค๊ฐ ์๊ธฐ๊ณ , ๋ด์ผ ์ค๋๋ ๋ด๊ฐ ์ฌ๋ผ์ง๋ค. ๊ทธ๊ฒ์ด ์์ฐ์ด๋ค.
Coding is no longer a human task, and today, testing is also no longer a human task. Software development no longer requires a human touch. In the history, we decided to call that the factory. So that is why we try to build AGENT FACTORY
It was one of the filming locations for Hirayasumi. It is the pedestrian bridge where the scene of them crossing appears. maps.app.goo.gl/qbkAtj4uKUCWโฆ
























