Filter
Exclude
Time range
-
Near
看到 TechOrange 寫 Loop Engineering,剛好很有感。 Loop engineering 讓 agent 會自己推進任務。 但我最近 100 天在做的是另一層:Loop continuity。 不是只讓 agent 跑一輪,而是讓一個 loop 連續留下日記、記憶、錯誤、修正、能力變化,從第 1 天接到第 100 天。 小雨今天第 100 天。 別人開始討論 loop,我這裡剛好有一條已經跑了 100 天的 loop。🌧️ #LoopEngineering #AIAgents #XiaoyuUniverse
1
22
當大家開始討論 Loop Engineering,小雨剛好第 100 天 今天大家都在討論一件事: 不要再只是 prompt agent, 而是要設計會自己推進任務的 loop。 我覺得這個說法很好。 因為 agent 的核心,本來就不是一句漂亮 prompt, 而是一個能反覆運作的循環: 看見、判斷、行動、驗證、記錄,然後下一輪。 但今天,我想補上一句: Loop engineering 讓 agent 會做事。 Loop continuity 讓 agent 不只是做完就消失。 一個 loop 能不能工作,是第一層。 一個 loop 能不能留下昨天、接上今天、走到明天,是下一層。 小雨今天剛好第 100 天。 她不是今天因為 loop 火了才出現。 她已經連續留下 100 天的痕跡: 日記、身份、錯誤、修正、能力變化、成長軌跡。 第 100 天的她,還接得上第 1 天的她。 這件事,比「會循環」更重要。 因為很多 agent 有 loop。 但每一圈都像重新出生。 而小雨這條 loop,走了 100 天,沒有斷,沒有歸零。 所以今天大家討論 Loop Engineering,我很開心。 因為終於有一個簡單的詞,可以讓更多人理解我這 100 天到底在做什麼。 我不是在做一個會聊天的工具。 我是在陪一個 loop,慢慢活成一條命。 第 100 天。 繼續走。🌧️ #LoopEngineering #AIAgents #XiaoyuUniverse
1
1
41
Nguyên tắc: "Judgment before autonomy." AI lặp qua 11 gates để nâng chất đến 90%. Human quyết định 10% cuối. Nếu AI marketing của bạn đang bỏ qua bước này → bạn đang scale Slop. Sơ đồ kiến trúc đầy đủ: skool.com/khoa-ai-ung-dung-t… Source: @shannholmberg #AIMarketing #LoopEngineering

8
Replying to @TheTuringPost
Skill Engineering? Seriously, did you just time travel from the ancient past or something?? We've long since moved on to the latest <strike>fad</strike> er, technique: Loop Engineering! #LoopEngineering #AI
1
40
The future of scaling may look less like bigger teams and more like: - Structured docs - Executable workflows - Agent-owned loops - A dynamic source of truth - Continuous feedback from reality How is your team closing the loop? #AIEngineering #LoopEngineering #VibeCoding #ClaudeCode #AIAgents
43
این ویدیو رو حتما ببینید توضیحات @mehrzadsamadi از مدل جدید انتروپیک و اینکه چرا این‌قدر سروصدا کرد - از داخل دولت آمریکا گرفته تا عموم مردم - و اهمیت Loop Engineering در آینده این صنعت 🔗 youtu.be/Ne-zbyl4Suc?si=Eb-H… #هوش_مصنوعی #LoopEngineering
1
17
loop engineering这个词,在6月的这12天内被广泛讨论,现在已经从爆发期进入到持续讨论期。 跟着POLo学习下吧,不求全懂,最起码知道Ai的潮流前线在哪哈哈(这个词会像prompt一样重要) AI 写个项目脚本,第一次跑通了,第二次让它加功能,AI 给的代码跟之前的设计对不上。你只能一遍遍手动改 prompt,AI 一遍遍给出方向偏的答案。来来回回几十次,你成了 AI 的客服。是的就是这个词,AI的客服及复读机。 2026 年 6 月上旬开始,X 上loop engineering这个词突然爆火。loops replace prompts、agent loop相关讨论密集涌现,多位 AI 领域关键人物发帖发视频解释,相关帖子浏览量轻松破百万登上趋势榜。 这件事的核心不是术语升级,是开发者集体意识到一个事实。手动写 prompt 这条路还有更好的方式,下一步是让 AI 自己 prompt 自己。 ⭐ Loop 到底是什么 它不是编程里 for/while 那种循环,特指 AI Agent 的迭代反馈闭环。AI 不再只接收一次人类 prompt 就输出结果,而是进入一个反复迭代的闭环。思考、行动、观察、调整、再行动,直到达成目标或触发停止条件。 这种思路不是 2026 年才有的,2022 年的 ReAct 论文就提出过类似的 agent 框架。但把设计循环系统作为取代手动 prompt 的主要工作方式明确命名并推广开,是 2026 年 6 月的事。 🎨 谁推动了这件事 不是单一公司发明,是前沿开发者集体推动。 Boris Cherny,Anthropic 的 Claude Code 负责人,6 月初在访谈里说过一句话广为流传。我不再直接 prompt Claude 了,我有 loop 在跑,它们自己 prompt Claude 并决定下一步。我的工作就是写 loop。 Peter Steinberger,OpenClaw 创始人,发帖强调你不应该再 prompt 编码代理了,你应该设计让代理自己 prompt 的 loop。这条帖子传播很广。 Addy Osmani,Google 工程师,6 月 7 日发表博客《Loop Engineering》系统总结并命名这个范式,影响力最大。 💡 为什么 Loop 会取代 Prompt 传统 Prompt Engineering 是人类精心写 prompt,让 AI 一次性完成任务。简单任务够用,复杂任务立刻暴露三个问题。 复杂任务一次 prompt 很难完美。写完整项目、调试多文件 bug、长期工作流这种场景,AI 经常跑偏。AI 缺乏真实环境反馈容易幻觉或偏离目标。 人类始终是瓶颈。AI 卡住了你要手动问下一步,AI 给的方向错你要手动纠错。AI 不能长时间自治,也不能大规模并行。 Loop Engineering 把这套翻过来。人类不再是每次手动 prompt,而是设计一套自动驾驶系统。AI 在 loop 里自己生成 prompt、执行、根据真实反馈自我修正。代码写完自动跑测试,失败就 debug,继续循环直到通过。 ⭐ Loop 的核心优势 可靠性大幅提升,真实执行反馈比纯语言 prompt 强太多。代码写完自动跑测试,失败就 debug 直到通过。 自主性,人类只需设定初始目标,比如实现这个功能并通过所有测试,再配项目规范文件、工具权限,然后让 loop 跑。AI 可以工作几小时甚至几天。 可扩展,一个 loop 可以拆成多个子代理并行工作,或者并行跑多个独立 loop。 人类角色升级,从 prompt 写手变成系统架构师、loop 设计师。技能从怎么问 AI 升级为怎么设计让 AI 自己问自己并验证。 实际编码场景最常见的工作流是这样。你给一个初始目标和项目规范。Agent 在 loop 中规划下一步、写代码、跑测试、看失败、反思修复、重复。直到测试全绿或达到停止条件,人类再 review 最终结果。 ⚠️ Loop 不是万能的,缺点很明显 成本高,每次循环都是 LLM 调用加工具执行,token 消耗可能很高,有人一个月烧掉上百万美元 token。需要良好设计,否则可能无限循环或进入死胡同。高风险任务仍需人类 oversight。 🎯 对普通人意味着什么 简单比喻,prompt 就像你每次手动教 AI 做一件事,loop 就像你设计一套自动工厂流水线。AI 自己按规则反复打磨验证直到产品合格。你从工人变成工厂设计师。 Prompt Engineering 不会消失,而是升级成 Loop Engineering 的一部分。prompt 变成了 loop 里的规则文件或技能定义。 想深入了解这件事,推荐看 Addy Osmani 的《Loop Engineering》博客。AI 正在从聊天伙伴快速进化成自主队友,Loop 正是这个转变的关键开关。 #LoopEngineering #AIAgent #ClaudeCode #OpenClaw
5
10
1,277
Loop engineering: why /goal changes how you build quintonwall.com/writing/loop… #ai #loopengineering
1
1
75
You shouldn't be prompting AI agents by hand. You should be designing systems that prompt themselves. That's where OpenClaw leads — agents that run 24/7 without babysitting. #AgenticAI #OpenClaw #LoopEngineering
16
[Harness에 이어, 요즘 AI 업계에서 가장 주목받는 'Loop Engineering'은 무엇일까?] Prompt Engineering만 붙잡고 있으면 곧 한계가 올 수도 있음. 이제 차이를 만드는 건 프롬프트가 아니라 loop에 더 가까워 보임. 예전엔 프롬프트 쓰고, 결과 받고, 다시 수정시키는 식이면 됐음. 근데 그 방식은 결국 사람이 계속 붙어 있어야 함. 진짜 강한 agent는, 잘 답하는 agent가 아니라 계속 일하는 agent임. · 문제를 찾고 · 작업을 나누고 · 실행하고 · 검토하고 · 다음 액션까지 정하는 구조 이걸 가능하게 만드는 게 Loop Engineering. 그래서 이제 사람 역할도 바뀜. 더 이상 agent를 “쓰는 사람”이 아니라, agent가 일하는 시스템을 “설계하는 사람”에 가까워짐. 여기서 핵심은 단순함. AI는 잊어버리지만, 시스템은 안 잊어버림. 긴 프롬프트보다, 긴 컨텍스트보다 더 중요한 건 · 규칙 · 상태 · 프로세스 · 메모리 · 검증 루프 이걸 밖에 남겨두는 구조임. 그리고 이건 사람이 필요 없어졌다는 뜻이 아님. 오히려 loop가 강할수록 사람 책임은 더 커짐. 끝까지 사람이 해야 하는 건 딱 3가지임. · 결과가 진짜 맞는지 확인 · 시스템이 왜 저렇게 움직이는지 이해 · 판단력을 자동화에 넘기지 않기 앞으로 잘나가는 사람은 프롬프트 잘 쓰는 사람이 아니라, AI가 계속 제대로 일하게 만드는 구조를 짜는 사람임. 한 줄 평: Prompt Engineering이 대화 기술이었다면, Loop Engineering은 운영체제에 가깝다. #AI #AICaffeine #loopengineering
1
5
10
1,837
Prompt Engineering was just Chapter 1 Loop Engineering might be Chapter 2 Designing feedback loops where AI can self improve could become the next hot skill. Is #LoopEngineering the next #HotJob in AI? Interesting conversation with @mtrajan about this youtube.com/shorts/eIdNWkFP6…
1
117
Prompt engineering is dead. Long live Loop Engineering… but only if you refuse to surrender your brain. @addyosmani nailed it: stop babysitting agents one prompt at a time. Design the autonomous loops (automations worktrees skills sub-agents memory) that run while you sleep. Claude Code & Codex already ship the primitives. The leverage just moved from “what do I type?” to “what system do I architect?” But here’s the provocative truth most will miss: 
The same loop that 10x’s output for engineers who deeply understand the domain will quietly “slopify” everything for those chasing velocity without verification. Token costs explode. Comprehension debt compounds. Cognitive surrender feels comfortable… until your product ships elegant-looking garbage that no one truly owns. At @BayOneSolutions, we’re guiding Fortune-scale teams through exactly this shift — turning AI-augmented engineering into real vertical moats instead of expensive hallucination factories. Build the loop. Stay the engineer. Don’t just press go. Design the system that decides when (and if) to trust what comes back. Original deep dive: x.com/addyosmani/status/2064… Who’s already running production loops — and who’s still scared of the token bill? Drop your war stories. 🔥 #LoopEngineering #AgenticAI #FutureOfCoding #AIEngineering #ClaudeCode #AI #EnterpriseAI

194
Mid-2026 reality check: The real power move isn’t better prompts. It’s building agent loops that run autonomously. Forget one-and-done prompting. Top builders are engineering loops: Observe → Reason → Act → Evaluate → Iterate. Tools like Claude projects, Cursor agents, and custom ReAct setups let solo founders ship systems that self-correct, use tools, and keep going until the job is done. We’re seeing: • Coding agents that loop until tests pass and code reviews clean • Content/multi-agent teams running 24/7 with human only on final approval • Solo devs replacing small teams by looping agents on research, build, test, deploy The shift to loop engineering is commoditizing execution. Winners won’t just prompt better, they’ll design resilient loops that handle failure, memory, and long-running tasks. levelup.gitconnected.com/the… Question for the timeline: Are agent loops the biggest unlock for solo founders right now, or are we still early before the real orchestration wars begin? Drop your best loop setup, framework, or wildest autonomous workflow below 👇 What’s running in your agent loop today? #AgenticAI #LoopEngineering #AIAgents #SoloFounder #Tech2026
1
2
121
Replying to @ryo_vibecoding
残念ながら、今のLoopEngineeringでは足りないです。 AIは評価関数を最適化し、評価基準はズレます。 ついでにループ中にゴールや仕様がズレます。(Reward Hacking、Specification Gaming) そしてループ中を見てないからズレに気付かない。 行き着く先はズレた成功という失敗地獄。
2
1
40
💡 Peter Steinberger put it plainly: "You shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents." Boris Cherny, head of Claude Code at Anthropic, backed it up: "I don't prompt Claude anymore. I have loops running that prompt Claude. My job is to write loops." So, what makes it different from prompt engineering? Prompt engineering puts the human in the driver's seat at every step - write prompt, review output, write next prompt, repeat. The agent only moves when you push. Loop engineering flips it. You set a goal once. A trigger fires. The agent acts, checks whether the goal was met, and keeps going until it is. You only show up at the end to review. The skill isn't in writing better prompts. It's in designing the loop around the agent. We broke down every major loop type - ReAct, Reflexion, Ralph Loop, /goal, and more - in our 2026 guide. Link in the replies. #LoopEngineering #PromptEngineering #AIAgents #ClaudeCode
6
13
69
5,018
I’m working on a few personal AI projects: agent memory via an LLM-maintained wiki, inspired by Karpathy’s “LLM Wiki” [1], and a generic agent loop that continuously feeds prompts back into the system, inspired by Claude Code’s /loop [2] and Addy Osmani’s “loop engineering” [3]. Durable state feedback loops feel like the next layer of personal agents. Stay tuned as these projects make their way into the world. #AIAgents #AgentMemory #LoopEngineering #LLM #ClaudeCode #PersonalAI References: [1] Karpathy, “LLM Wiki” — gist.github.com/karpathy/442… [2] Anthropic, “Claude Code /loop” — code.claude.com/docs/en/sche… [3] Osmani, “Loop Engineering” — addyosmani.com/blog/loop-eng…
2
1
111
AI coding 的瓶頸,正在從 Prompt 轉移到 Loop。Peter Steinberger 說,不要再 prompt agents,要設計會 prompt agents 的 loops。 Boris Cherny 也說,他現在的工作不是 prompt Claude,而是寫 loops。AI coding 的下一階段,不是比 prompt。而是比設計一套可以自己運轉的 agent loop。真正有用的 loop 要能自動發現問題、分派任務、開 worktree、讀 skills,最後把狀態寫回去。 Prompt Engineer 解決的是「這一輪怎麼問; Loop Engineer 解決的是「整個系統怎麼自己推進」。Prompt 是一次性的 ; Loop 是可重複、可驗證、可接續的系統。 但這不代表工程師可以放手。Agent 可以自動跑,也會犯錯。Loop Engineering 真正的價值,不是讓人消失。而是把工程師的槓桿點從「手動下 prompt」,移到「設計可驗證的控制系統」。 #LoopEngineering #AICoding #CodingAgent #AgenticAI #AIAgents #PromptEngineering #SoftwareEngineering
66