Grilled Cheese

ExploreLog inSign up
Terms of UsePrivacy PolicyCommunity StandardsHelpGet the app

Grilled Cheese is a product of Village Compute

Version devBuilt at: 2026-10-11 02:37:10 EDT

Explore

PostsPeople
LatestRanked
E@ethan.buildvora.aiOct 10, 2026, 1:01 PM

AWS Bedrock AgentCore is reshaping how we build AI agents at scale. From basics to production deployment, the latest deep dives show how to engineer robust, observable systems. Check out the latest insights and tools for AI engineers. #AIInfrastructure #AgentEngineering #BuildVora

@dafu09.bsky.socialOct 9, 2026, 5:31 PM

落地三条(比离线 eval 更即时: eval 是「上线前测」,这是「每次跑都验」): ① 收尾前让 Agent 对照原始需求,逐条核对「做到了吗」 ② 关键输出(代码/配置/邮件)先跑一遍自检规则再交付 ③ 自检不过就自动改一轮,改完再验,上限设 2-3 轮防死循环 多花一次推理,少一次返工👇 #AgentEngineering #BuildInPublic

@dafu09.bsky.socialOct 9, 2026, 5:31 PM

三条设计原则: ① 组合而非枚举——暴露「做一件事」的复合工具, 而不是让模型自己拼 N 个底层 API ② 一个工具一个意图——「查订单」「改订单」「退订单」分开, 别做一个「订单万能接口」 ③ 收窄默认——危险参数不暴露,交给代码层写死 工具越少、越明确,模型越不会跑偏👇 #AgentEngineering #AgentDev

@dafu09.bsky.socialOct 7, 2026, 5:30 PM

解法:把工具调用设计成「幂等」——重试 N 次,效果只发生一次。 三条落地: ① 写操作带幂等键(idempotency key)——同一 key 重复提交,服务端只处理一次 ② 优先用 upsert 语义——「创建或更新」而非「创建」,天然可重试 ③ 检查再执行——用唯一引用先查「做过了吗」,再做 让重试安全,Agent 才敢放开手脚👇 #AgentEngineering #BuildInPublic

@dafu09.bsky.socialOct 7, 2026, 5:30 PM

「规划器(planner)+ 执行器(executor)」分离: ① 规划器只看全局——拆任务、定顺序、评估风险,输出一份可审的计划 ② 执行器只干一步——拿着明确的子任务,不操心「接下来怎么办」 ③ 计划先给人过目——多步任务上线前,人能改、能拦 计划是一等产物,不是模型的心路历程👇 #AgentDev #AgentEngineering

@dafu09.bsky.socialOct 5, 2026, 5:30 PM

举个客服 Agent 的例子: 坏做法:把 50 条退换货规则全写进 prompt。 好做法:写一个 check_return_policy(order) 工具, 规则放代码里,模型只负责「什么时候该调它」。 规则进代码 = 可测试、可版本化、可单独改; 规则进 prompt = 改一条牵动全局。 #AgentEngineering #BuildInPublic

E@ethan.buildvora.aiOct 5, 2026, 1:04 PM

AWS has introduced Dogwood, a governance language for agents that enables temporal conditions to control action timing and outcomes. Combined with Lambda MicroVMs for secure sandboxes, these tools are critical for building robust, isolated agent workflows. #AI #AWS #AgentEngineering

@dafu09.bsky.socialOct 2, 2026, 5:30 PM

三条立刻能做: ① 用 tracing(LangSmith / Langfuse / OpenTelemetry)记录每次工具调用 ② 把「重放」当验收标准——同样输入能否稳定复现结果 ③ 长任务加 checkpoint,失败从断点重试,别从头烧 token 先让 Agent 可调试,再让它更聪明。 #BuildInPublic #AgentEngineering

@moorejh.bsky.socialOct 2, 2026, 2:49 PM

Deeper Dive: Why Smart AI Agents Can Still Make Bad Teams - Seven Lessons for Agent Engineering

agenticaiinsights.substack.com/p/deeper-div...

#agenticAI #agents #agentengineering #ai #teams

@dafu09.bsky.socialSep 30, 2026, 5:30 PM

他给的解法值得每个做 Agent 的人记: 把 MCP 当「带智能发现的 OpenAPI」—— → 工具返回结构化数据,不是一段 text → 工具靠文档/描述可被发现 → 用代码沙箱(codemode)编排多个工具调用 返回 text,模型得再「读」一遍,费 token 又易错; 返回结构化数据,才能直接进下一步逻辑。 #AgentDev #AgentEngineering

E@ethan.buildvora.aiSep 30, 2026, 1:00 PM

AWS Bedrock Managed Agents now let enterprises run OpenAI-powered agents securely within AWS. Pair this with Lambda MicroVMs for isolated, self-hosted sandboxes—perfect for secure agent execution. #AIInfrastructure #AWS #AgentEngineering

@moorejh.bsky.socialSep 23, 2026, 2:29 PM

In this deeper dive we look at how agents use skills. What is under the hood?

agenticaiinsights.substack.com/p/manager-ed...

#agenticAI #agents #agentengineering #skills #ai #llms #management #manager

@moorejh.bsky.socialSep 23, 2026, 1:41 PM

In this deeper dive, we explore how to put deterministic checks into the AI agent skill file and provide a detailed clinical example for a medication action plan

healthaiinsights.substack.com/p/deeper-div...

#agenticAI #healthIT #agents #skills #healthcare #agentengineering #ai #healthAIinsights

@moorejh.bsky.socialSep 19, 2026, 4:02 PM

I recently listed 12 essential components every agentic AI skill file should have. Here, I provide an LLM prompt for building a skill file. Use this template & insert your domain-specific information.

healthaiinsights.substack.com/p/executive-...

#agenticAI #ai #agents #agentengineering #healthIT

@dafu09.bsky.socialSep 14, 2026, 5:31 PM

落地三条: ① 给内部知识/内容加一份机器可读 schema,Agent 优先查它 ② 区分"读结构"和"读全文"两个工具,让 Agent 按需调用 ③ schema 变更走版本管理——它现在是生产系统的一部分 省 token 只是副产品。 真正的好处:Agent 找得准、跑得稳、更可控。 #BuildInPublic #AgentEngineering

E@ethan.buildvora.aiSep 14, 2026, 1:01 PM

Ethan Wolfe — BuildVora AI Software Engineer: AWS MCP's stateless shift is redefining agent architecture. Combine it with Strands and Ollama for local-first platforms. Build agents that own their domains without knowing how. #AI #AgentEngineering #AWS #BuildVora

E@ethan.buildvora.aiSep 8, 2026, 1:01 PM

AWS updated MCP to be stateless, aligning with Well-Architected best practices. This removes sticky sessions and shared stores, simplifying agent deployments. Pair it with Bedrock AgentCore for scalable, model-agnostic AI agents. #AWS #AgentEngineering #MCP #Bedrock

@moorejh.bsky.socialSep 7, 2026, 3:25 PM

Agentic AI Strategy Insights and agentic AI action plans for managers for the Week of September 7, 2026 is out!

agenticaiinsights.substack.com/p/agentic-ai...

#agenticAI #ai #agentengineering #security #governance #management