Daniel Lee retweetledi
Daniel Lee
6.7K posts
Daniel Lee
@dylayed
개발 이야기를 좋아합니다.
San Francisco, CA Katılım Eylül 2015
2.2K Takip Edilen11.3K Takipçiler
개인적으로 가장 주의깊게 보고있다.
개발조직이 강한 회사라면 background agent 정도라면 뚝딱 만들 능력을 가진 사람이 많고, 또 잘 만들었을 경우 사용 지표가 굉장히 좋을테니 윈윈 하는 경우가 많을것.
결국 쿠버네티스 같은 하나의 oss 구현체로 통일될까? 클로드나 코덱스가 결국 먹을까?
Utpal Nadiger@utpalnadiger
Everyone is building internal agents! @WorkOS built Horizon @sentry built Junior @tryramp built Inspect @coinbase built Mux @browserbase built bb @mintlify built kb @harvey built spectre @stripe built minions Everyday, we're seeing new orgs spinning up internal background agents, collaborative and in the cloud, a large % of them driven from slack. Adoption is skyrocketing - the image below is a graph of how adoption is growing at ramp. Stuff like customer support, sre/incident response, even context gathering for sales is reaching (almost) full autonomy, allowing the humans in these orgs to provide the company more value outside of largely deterministic busywork. We saw that this was one of the fastest growing use cases at @opencomputerhq and shipped our durable agent sessions api to make a fully managed, cloud agent, an API call away. Give it a shot? We're really hungry for feedback. Docs: docs.opencomputer.dev/agent-sessions… Playground: managedagents.sh
한국어
@sqs Wow I wonder if chatgpt subscription has something to do with the "everyone" usage being dominated by Medium
English
We now publish live % usage of Amp mode, for everyone and for just the Amp team.
ampcode.com/models
English
@nacyotKim 낙교님이 올리신 시간을 함축한다는 맥락도 생각나더라구요. 아직 이 새로운 초능력을 어떻게 사용하는지 모른다는 생각이 듭니다
한국어
@dylayed 저도 비슷하게 느낍니다. 너무 개인한테 시선이 가있는데, 사실 지금 차별화는 개인이건 팀이건 바닥을 다지는 데 있는 것 같기도 하고요. 다른 얘기지만, AI 이후에 잠자고 있던 HomeLab의 인프라가 급속도로 발전하고 있습니다 😂
한국어
"AI로 생산성이 얼마나 올랐나요?" 라는 질문에 우리 조직 리더의 대답을 마음에 머금게 된다.
"개인 속도가 몇 배 빨라지는게 핵심인가? 내가 느낀 변화는 손대기엔 너무 벅차고 자동화하기엔 애매해 '방치했던' 조직의 문제들을 마침내 해결할 수 있게 됐다는 것이다.
한국어
그냥 개인의 생산성 향상이 아니라, 하나의 개발 조직이 현실적으로 할 수 있는 일의 경계선이 바뀌고 있다."
듣다 보니 Bun의 Rust rewrite이나 Anthropic의 Project Glasswing이 떠올랐다.
나는 아직 자동차가 막 발명된 시대의 마부처럼 느껴진다.
한국어
@nicolaygerold Really like this simplification.
One thing I noticed for myself is that ever since I hooked up ChatGPT subscription to Amp, I end up preferring modes that primarily used GPT models (to save $$). And I see myself recalibrating again here to see which modes are primarily GPT...
English
Agent modes in Amp are now a dial.
low → medium → high → ultra
The old question was “which model/prompt/effort combo do I need?”
The new question is: how hard is this task?
English
Daniel Lee retweetledi
No not at the moment. If you’re running `amp —no-tui` on localhost you can obviously just pull that up. If you’re driving it from another machine, you could have it use ngrok. But at the moment we don’t have a first class way to expose ports on arbitrary machines to public internet
English
@dylayed It is a remote agent? Or do you mean you want to bring your own dev environment / sandbox? (In which case the answer is this: ampcode.com/news/agents-in…)
English
Waiting to see what @AmpCode team thinks about this model 👀
Cursor@cursor_ai
We've partnered with SpaceXAI to train Grok 4.5. It’s our most powerful model yet and the first we've built for more than software engineering.
English
If Amp had a subscription, what should it include and how much should it cost?
English
@thorstenball All I want now is a native iOS app 🙏
PWA is good, but buggy at times, e.g. can't open the left sidebar and must stare at the amp logo animation. It's beautiful but I would also like to get back to my session.
English
You can now remotely launch Amp agents anywhere you can run amp.
That's right: anywhere.
Laptops, cloud dev boxes, production servers, Raspberry Pis, Mac Minis, fridges and lawn mowers and dish washers.
If it can run amp, you can remotely start agents on it.
English
@davidcrawshaw @ssh_exe_dev Took Shelley around for my long weekend and wow it's a great harness. Having a native iOS app was cherry on top! Kudos.
English
@dylayed @ssh_exe_dev I do yes. Some others here do too, some have written their own orchestrator / factory systems. I find Shelley works really well, though my own orchestrator is definitely coming.
English
More details about using your ChatGPT subscription in @ssh_exe_dev: blog.exe.dev/chatgpt-subscr…
English
개인적으로 Fable 마케팅은 거의 완벽하다고 본다.
-너무 강력해서 미국 정부가 막았음!
-근데 다행히 안전하게 쓰는 법을 찾아냄. 진짜 다행임
-원래 엄청 비싼데... 에라이 모르겠다 1주일 무료 체험 제공!
-별로 대단한지 모르겠다면? 그건 니가 쓸줄 몰라서 그래.
Thariq@trq212
한국어
@stevekrouse I feel the same way, but I'm also wondering if the workflows that uses large number of subagents super aggressively are working better with Fable as the brain.
English
Been using Fable all morning. Not impressing me. Feels like Opus.
I gave it the task to port an eval – replit's vibench – to run on Val Town. Seems well-scoped, right?
But it's constantly making errors, needing me to redirect it
I want to get the hype. What am I doing wrong?
English