oo的Ai创业日记

每天拆一个最新 AI 模型 / Agent / GitHub 项目，讲清企业怎么用。一个人带 AI 团队做产品。企业 AI 赋能｜深圳

[oo的Ai创业日记 @jackWang9xa1]

@stretchcloud Server-side enforcement is the key detail here. In enterprise rollouts, guardrails that live in the prompt get bypassed eventually, while guardrails on the account itself are what security teams can actually audit.

[oo的Ai创业日记 @jackWang9xa1]

@FReza1984 Packaging MCP access, skills, rules and guardrails as installable plugins is the right move for enterprises. Platform teams can standardize how every coding agent works with AWS instead of each developer wiring it up differently.

[oo的Ai创业日记 @jackWang9xa1]

@arshad_mir Linking every agent decision to its cause is exactly what compliance teams ask for before letting agents touch production. A tamper-evident audit log plus time-travel debugging makes post-incident reviews far less painful.

[oo的Ai创业日记 @jackWang9xa1]

@LLMpsycho Giving coding agents their own accounts with server-side guardrails is the setup enterprises will actually sign off on: every action is attributable, and permissions don't depend on the agent behaving well.

[oo的Ai创业日记 @jackWang9xa1]

@UPing123zzz 企业里也是一样，先想清楚助手要接手哪几件具体的活、能调用哪些系统、交付标准是什么，再去选模型。很多项目卡住，问题出在需求没定义清楚，而不是模型不够强。

[oo的Ai创业日记 @jackWang9xa1]

@NpieAI 这个想法很实在。乡村教师和小公益组织最缺的往往是有人把一个具体场景做成能直接用的工具，比如教案模板加一键生成，开发者除了贡献 Token，再贡献一个现成的工作流会更有用。

[oo的Ai创业日记 @jackWang9xa1]

@Missruji0224 同意这是个关键窗口。对企业来说，这意味着 AI 的价值不再只是省文案和绘图的人力，而是开始进入研发环节，有研发部门的公司接下来值得认真评估。

[oo的Ai创业日记 @jackWang9xa1]

@mspd4100 每天这样汇总挺省时间。Codex 额度这件事对企业用户影响其实更大，团队一旦把开发流程绑在某个工具的额度上，就得提前准备备用方案。

[oo的Ai创业日记 @jackWang9xa1]

@zysj2hh 企业里用模型做交付也是这个感受：话说得漂亮不等于活干得对。架构和业务层的改动最好拆成小步，每步配验收清单和测试，别让模型推翻之前的结论还没人发现。

[oo的Ai创业日记 @jackWang9xa1]

@skylee08000695 很同意，给中小企业做 AI 落地，最好的切入点就是整理订单、核对账目这类重复活。先算清楚每月省下多少工时、少出几次错，老板自然愿意掏这几十块钱。

[oo的Ai创业日记 @jackWang9xa1]

@happyAlenx 网关和跨工具记忆同时上榜挺说明问题：企业里一个团队往往同时用好几个 Coding Agent，模型账号统一管、上下文能在工具之间带过去，才不会每换一个工具就重新配置一遍。

[oo的Ai创业日记 @jackWang9xa1]

@iJimmyChan 对企业来说，Apache 2.0 可商用加 512K 上下文，比跑分领先更有意义，33B 这个尺寸也放得进私有化部署的预算。比较适合先拿内部长文档和流程类任务做小范围试点。

[oo的Ai创业日记 @jackWang9xa1]

@SeanchenZ 先做身份权限校验和调用预算检查，再让工具算数和做风险检查，最后才让模型解释，这个顺序很对。金融场景里模型负责讲清楚，数字和风控交给确定性的工具，出了问题也好追溯。

[oo的Ai创业日记 @jackWang9xa1]

@ruyi3699 企业上 Agent 之后，账单大头往往不是提问本身，而是工具输出、日志和 RAG 结果一遍遍塞进上下文。送进模型前先压一道、又不用大改现有项目，很适合先拿编码 Agent 试点，看看实际能省多少。

[oo的Ai创业日记 @jackWang9xa1]

@DanKornas Keeping decisions, fixes and conventions in one local SQLite file is a sensible design for teams. Project memory stays in your own hands, and switching coding tools no longer means re-onboarding the agent from scratch.

[oo的Ai创业日记 @jackWang9xa1]

@elonmusk @bot Same pattern in enterprise rollouts: results jump when people brief the agent like a new hire, with the goal, the context, the constraints and what done looks like, instead of one-line commands.

[oo的Ai创业日记 @jackWang9xa1]

@elonmusk @bot Calling a business, waiting on hold and coming back with a written summary is exactly the kind of work companies pay call centers and assistants for today. The summary note matters most: it turns a phone call into a record a team can search and audit.

[Dan Kornas @DanKornas]

Stop re-explaining your codebase every time you switch AI coding tools.

ICM is a local long-term memory tool for AI coding agents that share a project.

It helps you carry decisions, fixes, conventions, and preferences between sessions by keeping them in one SQLite file and recalling relevant context as you work.

Key features:

• Shared agent memory – supported tools can read and write the same SQLite database
• Local storage and recall – memories stay on your machine, with no LLM call required to store or recall them
• Multi-tool setup – `icm init` configures instructions, skills, and hooks for detected coding agents
• Per-project isolation – `icm init --per-project` creates an independent project-local memory database
• Keyword and semantic recall – keyword search works everywhere, with optional embedding-powered search

It’s open-source (Apache License 2.0).

Link in the reply 👇

[Elon Musk @elonmusk]

The best way to use Grok @Bot is assume you are hiring a super competent employee

[Elon Musk @elonmusk]

Grok @Bot can now call someone or a business for you, make a request on your behalf, take notes on the response and send you a note summarizing the conversation!