AITOPAI热报
How-to

Warp 的 Zach Lloyd 演示：智能体连续失败 37 次后主动提交 PR 修复自身工作流

AI Engineer@aiDotEngineer
Post
View the post
Why we picked it

Warp 的 Zach Lloyd 放出的演示挺有意思：智能体同一错误摔了 37 次，干脆自己提了个 PR 修工作流。做 Agent 的朋友可以看看这种"失败次数触发自我修复"的设计思路，他这周还会在旧金山 AIE CODE 现场讲。

Warp 的 Zach Lloyd 在 X 上分享了一个智能体自我改进的案例：同一个失败重复出现 37 次后，智能体没有继续硬试，而是向它所在的工作流仓库提交了一个 PR，用来修复导致反复失败的流程本身。

Image · AI Engineer
Body

这个演示的核心是把"重复失败"当作信号——智能体检测到同一错误模式高频出现，就升级为对工作流定义的修改请求，而不是无意义地重试。Zach Lloyd 将在 11 月 10 日至 12 日于旧金山举行的 AIE CODE 大会上进一步讲解这套方法。

对构建智能体系统的开发者来说，这展示了一种让智能体从"执行者"变成"流程维护者"的具体模式。

Original · AI Engineer

The same failure, 37 times. Then a proposed PR to fix the factory. @zachlloydtweets of @warpdotdev shows how agents propose improvements to their own workflows. See Zach at AIE CODE this week. Nov 10 to 12, San Francisco. Apply: ai.engineer/code

Industry Coordinates & Source Verification

This report is categorized under AITOP's "How-to" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include 智能体, Warp, 工作流, AIE CODE.

Source attribution: curated from AI Engineer (@aiDotEngineer).

#智能体#Warp#工作流#AIE CODE

Keep reading

Recent stories with the same tag or section
ProductsBuildersMulti-source07:22
Chinese contentGoogle Antigravity 推出面向知识工作的新界面布局

Google 给 Antigravity 加了个适合写东西、做研究的界面，去掉了终端和 git 那些开发专属的东西，但底下的 agent 和 MCP 工具都是同一套。如果你不是程序员但想让 agent 帮你查资料、写文档，这个比 Cursor 更对味，设置里切一下就能用。

#Antigravity#Google#MCP/工具#智能体
Event
Varun Mohan@_mohansolo9 sourcesSource
ModelsUS vs ChinaMulti-source91°07:20
Chinese contentAnthropic 发布 Claude Fable 5.2 与 Mythos 5.2，8 天协调千个智能体造出原型操作系统

Anthropic 把旗舰模型价格砍了 20%，还让模型指挥一千多个子智能体 8 天写出能启动的操作系统。比起 OpenAI 的旗舰，它在长程任务协调和科学形式化证明上打法不同。做智能体编排和科研工具的开发者可以重点看协调机制的细节。

#Claude Fable 5.2#Anthropic#Claude Mythos 5.2#推理模型
Event
IT之家11 sourcesSource
ModelsBuildersMulti-sourceSelected06:52
Muse Ember Matches GPT-6 Luna on Browser-Use Benchmark at Lower Cost

Pricing per task, not raw benchmark scores, is becoming the deciding metric for agent builders, and Muse Ember now defines the cost-performance frontier for browser automation. Teams running web agents at scale can cut per-task cost by 2.4x versus GPT-6 Luna medium with no score loss on hundreds-of-steps workloads. Engineers evaluating browser-use stacks should re-benchmark their model choices against sub-5-cent tiers before defaulting to frontier models.

#Muse Ember#GPT-6 Luna#browser-use#agents
Event
Browser Use@browser_use11 sourcesSource
ModelsBuildersMulti-sourceSelected06:35
Muse Ember: Alexandr Wang Announces Small Efficient Agent Model

The pricing claim — a third of the cost of Haiku-class models — is the lever that could shift how teams budget for always-running agents. Engineering teams running high-volume agent loops should re-run their cost benchmarks against Muse Ember before renewing model contracts. No independent benchmark data is available yet, so treat the intelligence-per-dollar figures as vendor claims until third-party evaluations appear.

#Muse Ember#Haiku#small models#agents
Event
AI SDK@aisdk11 sourcesSource