自动驾驶公司

今天在《AI每日简报》中，我们聊聊AI如何推动自动驾驶公司的诞生。

Today on the AI Daily Brief, what it means for AI to create self driving companies.

《AI每日简报》是一档每天更新的播客和视频节目，聚焦AI领域最重要的新闻和讨论。

The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

好了，各位。

Alright, friends.

在正式开始之前，先说几件小事。

Quick announcements before we dive in.

首先，感谢今天节目的赞助商：Robots and Pencils、Blitzy、Section 和 Airtable。

First of all, thank you to today's sponsors, Robots and Pencils, Blitzy, Section, and Airtable.

想收听无广告版本？请访问 patreon.com/aidailybrief，或者在 Apple 播客上订阅。

To get an ad free version of the show, go to patreon.com/aidailybrief, or you can subscribe on Apple Podcasts.

如果您想了解如何赞助本节目，请发送邮件至 sponsors@aidailybrief.ai。

To learn more about sponsoring the show, send us a note at sponsors@aidailybrief.ai.

这期节目是周末特别版，主打长文深度内容，因为 Replit 的首席执行官刚刚发布了一篇重磅文章。

This is, of course, a weekend episode, a big thing slash long reads episode, and the CEO of Replit just dropped a great big thing piece.

那我们就开始吧。

So let's dive in.

今年一直在发生的一件事是，AI的发展已经不再只是让我们重新思考个人或团队的工作方式，而是开始从根本上改变公司的整体架构。

One of the things that is happening, and really has been happening all year, is that AI has advanced in such a way that it's no longer just making us rethink how we do work on an individual level or even a team level, but is starting to implicate the entire design of the company itself.

今年人们特别关注的一些实验，比如Pulsia，它最初是为完全无人公司设计的框架，现在已演变成一套全新的AI原生公司操作系统，能大幅降低创建和维持一家公司的启动门槛，哪怕公司里依然有人参与。

Some of the experiments that people have been really excited about this year are things like Pulsia, which started as a framework for entirely no human companies, and that has evolved a bit to be a new AI native company operating system that dramatically minimizes the base activation energy needed to build and maintain a company, even if it doesn't require there being no humans at all.

但不仅仅是这些全新的尝试，才体现出新型智能代理工作方式的影响。

But it's not just totally new efforts that are seeing the impact of new agentic ways of working.

事实上，尽管这些新兴创业项目听起来很酷，但它们的实际应用对现有企业来说，有时还是显得有点遥远。

In fact, as cool as some of these brand new startup style experiments are, ultimately, their examples can be a little bit hard to see how they apply to today's existing companies.

所以，当一些规模稍大、哪怕仍被看作初创公司的企业，开始分享他们真正行之有效的全新工作方式时，就特别值得留意。

Which is why it's really interesting to see when companies that are at least slightly larger, even if one might still consider them startups, are sharing their totally new ways of working that are, well, working.

上周，Replit的首席执行官阿姆贾德·马萨德发布了一篇题为《自动驾驶公司》的文章，就是一个绝佳的例子。

We got a great example of this last week when Replit CEO Amjad Masad published a new post called The Self Driving Company.

我先读一下他在X上发布的这篇博文，之后我们再一起探讨它的深层意义。

I'm going to read that blog post from X, and then we'll come back and talk about some of the implications.

自驱动公司：我们正开始看到，当一家公司学会自我运转时会发生什么。

The self driving company We're beginning to see what happens when a company learns to operate itself.

过去六个月里，Replit 的工程师们将代码产出几乎提升了两倍。

In the past six months, engineers at Replit have nearly tripled code output.

代码审查时间保持稳定。

Review times held steady.

代码回滚和产品事故数量也没有增加。

Reversions and product incidents have stayed flat.

质量指标有所提升，发布频率也加快了。

Quality metrics improved, and releases have accelerated.

原本预期的那些权衡取舍，一样都没出现。

All the typical tradeoffs you might expect have not occurred.

代码只是表象，真正有意思的是背后发生的变化。

While the code is the visible part, what's happening under the surface is much more interesting.

现在，智能代理能自动排查线上故障、审查代码提交、回答问题、分析业务数据、处理客服工单、调研销售线索，还能优化驱动 Replit Agent 本身的系统。

Agents now investigate production incidents, review pull requests, answer questions, analyze business data, triage support tickets, research sales accounts, and improve the systems that power Replit Agent itself.

感觉就像有一个统一的智能贯穿每个员工，但实际上并不是这样。

It feels like a single master intelligence threaded through every employee, even though it is not.

这是一套不断扩展的智能代理系统，遍布整个公司，接收人的目标，收集上下文，执行任务，检查结果，并在需要人类判断时主动上报。

It is an expanding system of agents operating across the company, taking goals from people, gathering context, performing work, checking the results, and escalating when human judgment is needed.

我们认为，这标志着一种新型组织的开端——自动驾驶公司。

We think this represents the beginning of a new kind of organization, the self driving company.

自动驾驶公司并不是没有人的公司。

A self driving company is not one without people.

人依然决定目的地。

People still choose the destination.

他们选择哪些问题重要，做出艰难的权衡，展现判断力，并对结果负责。

They decide which problems matter, make difficult trade offs, exercise taste, and take responsibility for the outcome.

但越来越多的时候，他们不再亲自完成每一步流程。

But increasingly, they do not perform every step required to get there.

这种转变从去年年底开始。

The shift began late last year.

和许多从事AI工作的人一样，我们过完圣诞节回来，感觉有些根本性的东西已经变了。

Like many people working in AI, we returned from the Christmas break feeling that something fundamental had changed.

模型现在能持续处理更长时间跨度的任务了。

Models could sustain work over much longer horizons.

以前总失败的任务，比如告警分类和根因分析，突然开始能正常运行了。

Tasks that had repeatedly failed, like alert triage and root cause investigation, began working.

AI开始解决我们一些最棘手的bug了。

AI started solving some of our most stubborn bugs.

我们不再把智能体当成只存在于编辑器或聊天窗口里的工具了。

We stopped treating agents as tools that lived inside an editor or chat window.

我们把它们细致地融入了公司运作的整个体系中。

We wove them carefully into the fabric of the company itself.

一旦工程团队证明了它的价值，推广就自然而然地蔓延开了。

Once engineering proved the value, adoption took on a life of its own.

一个接一个团队开始把最枯燥的工作交出去，腾出时间来做真正推动业务的战略和创意工作。

Team after team started offloading their most tedious work, reclaiming time for the strategic and creative thinking that actually moves the business.

人们并不觉得被自动化取代了。

People don't feel like they've been automated.

他们感觉自己被提升了。

They feel like they've been promoted.

这就是AI如何彻底改变了Replit公司工作方式的故事。

This is the story of how AI has completely changed the way that we work at Replit.

首先是工程团队感受到了影响。

Engineering saw the impact first.

在一月底，我们升级了基础设施，以便快速测试内部智能代理的应用场景。

In late January, we turned up infrastructure to experiment with internal agent use cases quickly.

我们利用了代理框架、微型虚拟机和远程文件系统，让任何工程师都能并行调度多个代理协同工作。

We leveraged our agent harness, micro VMs, and remote file systems infrastructure so any engineer could orchestrate swarms of agents in parallel.

之后，我们为整个系统设置了访问权限、令牌代理、审计日志和零信任网络防护。

Then we locked the whole thing behind access policies, token proxies, audit logging, and our zero trust network.

到那时，我们才放心让代理访问所有过去我们完成工作所需的工具：GitHub、GCP、Linear、Notion、Slack、Zendesk，还有更多。

At that point, we felt safe giving the agent access to all the things we used to get our jobs done: GitHub, GCP, Linear, Notion, Slack, Zendesk, and more.

有了跨系统的上下文，我们的工作效率有了巨大提升。

With context across systems, we saw a leap forward in productivity.

以前总失败的实验，现在变得轻松多了。

Experiments that previously failed became easy.

最直接的影响体现在代码统计上。

The most immediate impact was in coding stats.

三月份发布Agent 4之前，我们正处在冲刺周，通常这时候代码量会大幅上升。

We were in a sprint week leading up to Agent four release in March where we typically see a big spike.

会议没了，目标明确，工程师们直接进入全力执行模式，每天工作长达十六个小时。

Meetings disappear, scope is known, and engineering shifts into pure execution mode, often for up to sixteen hours per day.

但这次不一样了。

But this time was different.

我们的效率曲线以前所未有的方式飙升，这完全归功于新内部智能系统的应用。

Our productivity curve bent upward in a way none of us had seen before, which can be traced to the adoption of our new internal agentic system.

从一月初到六月底，贡献的代码行数增长了5.8倍。

From early January to late June, there was a 5.8x increase in the lines of code contributed.

这部分增长部分要归功于我们招到了合适的人才。

Part of this increase can be attributed to hiring well.

我们的新助手能让人更快上手，这当然很好，但为了更清晰的数据，我们可以排除掉招聘带来的影响。

Our new agent accelerates time to productivity, which is great, but we can remove the hiring effect for cleaner data.

保持同一组开发者不变，他们产出的代码量是以前的近三倍。

Keeping a consistent cohort of authors, we see 2.9x as much code as before.

过去，团队扩大时，如果每位工程师的产出能保持稳定，就已经算非常出色了。

Traditionally, it's considered excellent if you keep output per engineer flat as you scale a team.

我们不仅把团队规模翻了一倍，还让每人产出翻了三倍。

We just tripled per engineer rate while doubling the team.

你可能会问，这么多新代码谁来审查？会不会在审查环节造成新的瓶颈？

Now you might wonder who is reviewing all this new code, and whether we've created a new bottleneck in the review process.

我们的代码审查周期并没有变长，主要因为我们让助手参与了代码审查。

Our code latency is flat, largely because we put our agent to work in reviewing code.

它现在能评估风险等级，只有在必要时才会通知人工复审。

It's now able to assess risk levels and only call in a second human reviewer when necessary.

这意味着人类审阅代码的时间节省了30%，而且还在持续增长。

That means 30% and growing of human PR review time has been saved.

随着我们的代理负责编写和审阅更多代码，大家可能会担心质量会不会下滑。

With our agent writing and reviewing more code, we should be worried about quality.

如果我们看代码合并后的回退率和上报的问题数量，趋势都保持稳定。

If we look at PR reversion rates and incidents opened, trends are flat.

这说明我们在相对层面上其实做得更好了。

This means we're actually improving on a relative basis.

其中一个原因是，这些流程也都得到了代理的协助。

One reason is that these processes are also agent assisted.

人类进行代码审查时，有个代理作为协作者，因此能发现更多漏洞。

Human code reviews have the benefit of an agentic co reviewer, so more bugs get caught.

问题排查过程中，代理会帮忙定位根本原因，所以平均缓解时间正在下降。

Incident investigations are assisted by an agent that attempts to find the root cause, so Mean Time to Mitigation is going down.

最终的检验标准是，新增的代码是否真正带来了实际的价值。

The final test is whether additional code inputs represent real value output.

归根结底，工程团队的核心是为用户交付功能。

At the end of the day, engineering is delivering features for users.

我们用Linear跟踪项目进度，这样销售和市场团队就知道什么时候该跟用户沟通新功能上线。

We track projects in linear so that sales and marketing teams know when to communicate with users about new features.

你可以看到，随着代码量增加，项目完成的速度明显提升了。

You can see the rate of project completion is sharply up along with our coding volume.

总结一下。

Conclusion.

一个能自主运转的工程团队，不仅能交付更多代码，还能同时提升质量。

A self driving engineering team can ship more while raising quality at the same time.

下一节。

Section.

我们的智能代理系统正在推动工程流程的大规模自动化。

Our agent of agents is enabling loop engineering at scale.

我们再深入看看，具体是怎么运作的。

Zooming in gives us an idea of what this looks like.

当工程师找到方法创建循环，派出一批代理去完成可验证的任务时，我们看到了最显著的改变。

When engineers find ways to generate loops, sending a fleet of agents off to complete a verifiable task, we see the most dramatic change.

每位员工都能使用一个管理型代理，它可以生成多个子代理，帮你自动协调循环运行的任务。

Every employee gets access to a Manager Agent that can spawn multiple agents, enabling orchestration of agents working in loops on your behalf.

这些循环让我们的代码审查图看起来格外独特。

Loops result in some very unique looking PR graphs.

一位工程师完成了我们长期停滞的CSS系统迁移，并分享了他的经验。

One engineer completed a long stalled migration of our CSS system and shared his learnings.

另一位工程师自动化了一次迁移，让我们得以实现产品的本地化。

Another engineer automated a migration that enabled us to localize the product.

还有人自动处理了那些反复失败的测试维护问题。

Yet another automated flaky test maintenance.

我们的CTO最终用一群代理攻克了我们最难的一个网络bug，问题出在PSC NFT关闭时。

Our CTO finally cracked one of our hardest networking bugs related to PSC NFT shutdown with a swarm of agents.

我们对什么是可能的，所有之前的假设都彻底改变了。

All of our assumptions about what is possible have changed.

最令人兴奋的自动驾驶案例来自我们的AI团队。

The most exciting self driving example comes from our AI team.

他们打造了一个持续学习系统，能分析用户反馈、提出改进方案，并通过基准测试和A/B测试来验证成效。

They built a continual learning system that analyzes user feedback, proposes improvements, and uses a combination of benchmarks and AB tests to validate the wins.

Replit Agent正在自我进化。

Replit Agent is self improving.

构建还是采购的讨论方式已经变了。我们的内部代理也彻底改变了团队对软件采购与自研的权衡。

Section The Build versus Buy Conversation Has Changed Our new internal agent also changed conversations about whether we build or buy software.

我们经常试用新的AI工具。

We regularly try out new AI tooling.

购买现成方案能让我们更快推进，但我们也一直在持续评估市场。

Buying solutions can help us go faster, and we also assess the market constantly.

但随着我们自研能力的提升，这类外部采购会越来越少。

But the more we build, the less of this we will need to do.

我们的内部代理现在表现已经超过了许多业内公认的领先产品。

Our internal agent now outperforms products we test that are seen as market leading.

我们直接停掉了七位数的SaaS产品，因为内部用Replit完全自建的应用更优秀，员工都自发转过去了。

We just turned a seven figure SaaS solution because our internal app, built entirely in Replit, was superior and employees had migrated over.

突然间，市面上的工具都感觉像是专门为我们量身打造的。

All of a sudden, tools feel like they are built for us.

我们对知识库的深度整合和定制化，让其他解决方案相比之下显得很粗糙。

The deep integration with our knowledge bases and customization we've done makes other solutions feel inferior.

更让我们意外的是，我们的内部代理在对比多个垂直领域专业工具时也胜出了。

What surprised us more was that our internal agent also beat out vertical specific products we evaluated.

有个帮工程师处理告警和排查故障的工具，效果和我们的差不多，但成本却是我们的十倍。

A tool to help engineers triage alerts and root cause incidents came back with similar quality but at 10x the cost of running it on our agent.

另一个做自动化渗透测试的工具，发现的漏洞比我们的还少，价格却贵了十倍。

A tool that runs automated penetration testing found fewer vulnerabilities than our internal version at 10x higher cost.

这两个内部版本都顺利上线了，大幅缩短了故障响应时间，也增强了关键系统的安全防护。

Both our versions were put into production with ease, reducing MTTM in incidents and hardening critical systems against attacks.

考虑到我们还在不断学习，模型也在持续进步，这显然只是个开始。

With how much we're still learning, and how models are improving, it's clear this is only the beginning.

哪个部分？

Section?

不只是工程，而是整个公司。

Beyond engineering and into the whole business.

一家自动驾驶公司不会只停留在工程上。

A self driving company doesn't stop at engineering.

Replit的每个部门都在发生变化。

Every function at Replit is changing.

使用范围很快从工程部门扩展出去，主要得益于Slack接口。

Usage spread quickly out of engineering, mostly because of a Slack interface.

公司其他同事看到工程师用我们的智能助手处理任务，也试着自己用。

The rest of the company noticed engineers tagging our agent with tasks and tried it for themselves.

最初最受欢迎的用法是提问。

Initially, the most popular use case was asking questions.

把我们的知识库和代码状态结合起来，任何人都能明确产品需求，不用等工程团队回复。

By combining our knowledge base with the state of the code base, anybody could clarify product expectations without waiting for engineering input.

这些员工随后就能跟进修改文案或文档。

Those employees could then fix copy or documentation as a follow-up.

这立刻提升了我们响应用户的速度。

It was an immediate boost in being able to respond to users faster.

但这只是个开始。

But that was just the beginning.

从那以后，公司各个部门开始陆续贡献新的功能和集成。

From there, contributions of new skills and integrations started to come in from all parts of the company.

第一个重大突破来自我们的数据团队。

The first big unlock came from our data team.

他们为助手搭建了一个数据仓库的语义层，让它知道哪些表是权威数据源，以及它们之间的关联关系。

They gave the agent a semantic layer over our data warehouse, so it knows which tables are sources of truth and how they relate to one another.

现在，Replit 的任何人都能提出商业智能问题，并获得可靠的答案。

Now anyone at Replit can ask business intelligence questions and get a reliable answer.

他们可以直接用实时数据生成图表和演示文稿。

They can build charts and presentations from live data.

数据团队现在能把时间花在深入解决最棘手的问题上，而不是应付各种请求。

The data team spends its time going deeper on the hardest problems instead of fielding requests.

最近，一位产品经理自己完成了复杂的上线分析，因为我们的智能助手能理解代码库中的事件、这些事件在客户数据平台中的表现，以及如何将它们与复杂的订阅状态关联起来。

Recently, a PM was able to self serve complex launch analysis because our agent understands events in the codebase, how they show up in our customer data platform, and how to join those with complex subscription states.

销售团队也获得了同样的效率提升。

Sales found the same leverage.

销售开发团队用这个助手来发现并完善符合产品资格的潜在客户，利用的是其他通用工具看不到的内部信息。

The sales development team uses the agent to find and enrich product qualified leads, drawing on internal knowledge that more generic tools can't see.

这样一来，外联沟通都更有背景依据了。

So Outreach lands with more context.

客户经理用它来准备客户对话，了解谁从中获得了最大价值、哪些项目最活跃，以及使用额度与合同条款的匹配情况。

Account Executives use it to prepare for customer conversations to understand who is getting the most value, what projects are most active, and how credit usage tracks against their contract.

所有这些信息最终都会整理成带有品牌风格的演示文稿，针对每个客户量身定制。

This is all then packaged up into branded slides customized to the account.

一个自动化的销售团队能与客户建立更多、更高质量的接触点。

A self driving sales team has more, higher quality touchpoints with their customers.

我们的市场团队可以用这个助手，仅凭一个指令，就根据工程和产品团队的对话与文档，从零起草产品规格。

Our marketing team can use the agent to draft product specs from scratch with a single prompt based on conversations and documents produced across engineering and product.

这样一来，他们不用参加每一个会议，也能更快启动发布工作，随时掌握最新进展。

This gives them the ability to start moving on launches sooner and stay up to date, without needing to be in every single meeting.

他们有了更多时间来做规划和创意，确保产品上线后能产生更大的影响力。

They have more time to plan and be creative, which will ensure our releases have greater impact when they are out in the world.

我们的客服团队给助手设定了排查问题和执行标准流程的技能。

Our Support Team gave the agent skills to investigate issues and follow standard playbooks.

它可以选择用我们标准的客服语气回复客户，也可以把问题连同调查摘要一并转交给工程团队处理。

It can choose to offer a response in our standard customer service voice, or escalate to engineering along with a summary of the ticket and investigation.

一个能自主运转的客服团队，处理那些需要人工介入的复杂工单，速度提升了60%。

A self driving Support Team closes the hardest tickets, those escalated to humans, 60% faster.

用户能更快回到开发工作中去。

Users get back to building sooner.

在每一个案例里，人都没有被取代。

In every example, the human didn't get automated out.

他们得到了晋升。

They got promoted.

自动驾驶让执行者变成领导者，真正脱颖而出的是那些关注成果、明确方向的人。

Self driving turns doers into directors, and the people thriving are the ones who think in outcomes and set directions.

这如今是最有价值的工作。

That is the most valuable work there is now.

接下来去哪儿？

Where to next?

提升我们的效率令人兴奋。

Making ourselves more productive is exciting.

但真正激励雷plit员工的，是让技术更加普及。

But what really motivates the people at Replit is democratizing technology.

我们希望把这种全新的工作方式带给所有用户。

We want to bring this new way of working to all of our users.

我们正在全力以赴，确保在部署这套系统时，能兼顾策略、权限、安全和成本控制，实现规模化落地。

We're hard at work making sure we can do this with the policy, permissions, security, and cost controls needed to deploy this at scale.

Replit 最活跃的用户是那些正在打造真实业务的创业者和企业用户。

Replit's most active users are entrepreneurs and enterprise users building real businesses.

自动驾驶功能需要可扩展的安全机制，才能满足这些用户的需求。

Self driving needs safety measures that can scale to meet those users.

我们目前正在全力开发这项功能，但结合以上所有进展，你不用等太久。

We're hard at work building that now, but given all the above, you won't have to wait long.

我每天在这档节目中，都在探讨人工智能潜力与实际落地之间的差距。

I cover the capability gap between AI potential and AI reality every day on this show.

大多数公司还在摸索该怎么起步。

Most companies are still figuring out how to start.

Robots and Pencils 已经开始上线并快速扩展了。

Robots and Pencils is already launching and scaling.

Agendaq 和生成式 AI 已经在大型企业中投入生产，几周内就能落地。

Agendaq and Generative AI in production at large enterprises in weeks.

AWS 高级合作伙伴的数量在过去一年里翻了一倍多。

AWS advanced tier pattern partner more than doubled in a year.

他们正在招人。

And they're hiring.

有五十个职位空缺。

50 open roles.

如果你意识到这个时代与众不同，想亲身参与其中，而不是旁观，那这值得你认真看看。

If you're someone who knows this moment is different, who wants to be inside it, not watching it, this is worth a look.

在 Robots and Pencils，最好的点子会胜出，团队也始终保持顶尖水准。

At Robots and Pencils, the best ideas win, and the team is purposefully kept super high quality.

这里会成为你日后回想起来，觉得这辈子最明智的决定。

This is the kind of place you look back on as the best decision you ever made.

去看看 robotsandpencils.com/careers 吧。

Take a look at robotsandpencils.com/careers.

周末就该做点随性 coding。

Weekends are for vibe coding.

现在把你的兴趣项目变成现实，比以往任何时候都容易，赶紧打开你最爱的随性开发工具吧。

It has never been easier to bring a passion project to life, so go ahead and fire up your favorite vibe coding tool.

但周一马上就到，等你反应过来，面前已经是堆成山的微服务、70年代的老式COBOL系统，还有一条比你退休典礼还久的工程路线图。

But Monday is coming, and before you know it, you'll be staring down a maze of microservices, a legacy COBOL system from the 1970s, and an engineering roadmap that will exist well past your retirement party.

所以你需要Blitzy——首个专为大型企业级代码库设计的自主软件开发平台。

That's why you need Blitzy, the first autonomous software development platform designed for enterprise scale codebases.

每个冲刺周期一开头就部署，让你的路线图推进速度提升五倍。

Deploy the beginning of every sprint and tackle your roadmap 500% faster.

Blitzy的智能代理会读取你的全部代码，规划任务，然后以计算速度交付超过80%经过验证、端到端测试、高品质的代码，几乎全程自主完成。

Blitzy's agents ingest your entire codebase, plan the work, and deliver over 80% autonomously Validated, end to end tested, premium quality code at the speed of compute.

几个月的开发工作，压缩到几天内搞定。

Months of engineering compressed into days.

周末继续用你最爱的氛围编程工具搞你的兴趣项目。

Vibe code your passion projects on the weekend.

周一上班，就把Blitzy带进办公室。

Bring Blitzy to work on Monday.

去看看为什么《财富》500强企业都信赖Blitzy来处理关键代码，访问blitzy.com。

See why Fortune 500s trust Blitzy for the code that matters at blitzy.com.

这就是section.com。

That's blitzy.com.

有个残酷的事实：你的公司很可能正在花成千上万甚至上百万美元购买AI工具，但这些工具的利用率却低得惊人。

Here's a harsh truth: Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized.

虽然有一半的企业都配备了AI工具，但真正用它们创造商业价值的只有12%。

Half of companies have AI tools, but only 12% use them for business value.

大多数员工还是只拿AI来总结会议纪要。

Most employees are still using AI to summarize meeting notes.

如果你在公司里负责推动AI落地，那你需要Section。

If you're the one responsible for AI adoption at your company, you need Section.

Section是一个帮助企业全面管理AI转型的平台。

Section is a platform that helps you manage AI transformation across your entire organization.

它指导员工掌握真实的业务场景，追踪谁在用AI创造实际价值，并清晰展示AI在哪些地方真正产生了效益，又在哪里被浪费了。

It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value.

结果呢？

The result?

你将从单纯部署工具，转变为推动可衡量的AI价值。

You go from rolling out tools to driving measurable AI value.

你的员工不再只做会议纪要，而是真正解决业务问题，而你也能证明投资回报。

Your employees move from meeting summaries to solving actual business problems, and you can prove the ROI.

别再猜你的AI投入有没有效果了。

Stop guessing if your AI investment is working.

访问 sectionai.com 了解 Section 平台。

Check out Section at sectionai.com.

那就是 sectionai.com。

That's sectionai.com.

本集《AI每日简报》由 Hyperagent 赞助播出，你可以管理一支团队协作的智能代理舰队。

This episode of the AI Daily Brief is brought to you by Hyperagent, where you run fleets of agents your team can manage together.

新用户可获得一千美元的推理算力。

New users get $1,000 in inference.

别再等着在电脑上手动触发本地代理或聊天流程了。

Forget local agents and chat workflows waiting on your laptop to be prompted.

Hyperagent 在云端部署全天候运行的智能代理，直接使用你们团队已经在用的工具完成实际工作。

Hyperagent deploys always on agents in the cloud, doing real work across the tools your team already uses.

市场团队的代理能将竞争对手的动态自动生成落地页。

Marketing's agent turns competitor moves into landing pages.

销售团队的代理负责完善潜在客户信息、撰写邮件，并更新客户管理系统。

Sales's agent enriches leads, drafts emails, and updates the CRM.

运营代理则跟进文书流程，实时追踪预算使用情况。

Ops agent chases the paperwork and tracks the budget.

每个代理都能访问共享的信息，并严格遵守你们设定的权限范围和审批规则。

Every agent has access to shared context and follows your rules about scope and approvals.

现在是时候引入那些像同事一样可靠的智能代理了。

It's time you add agents that feel like teammates.

立即在 Hyperagent 上雇佣你的智能团队，由 Airtable 原班人马打造。

Hire yours at Hyperagent, built by the team at Airtable.

前往 hyperagent.com/aideallybrief 领取你的 1000 美元算力额度。

Claim your $1,000 in inference at hyperagent.com/aideallybrief.

好的。

Alright.

我觉得这不仅是一个非常有趣的案例，更进一步说，它提供了一种全新的思维方式，帮助我们重新理解公司该如何构建。

So I think this is both a super interesting case study, but also even more than that, frankly, a really interesting new mental model for how to think about company building.

接下来，我们聊聊几个关键的启示，特别是对你这样在思考如何让公司更自动化的人来说，能从这篇内容中带走什么。

So let's talk about a number of different takeaways and things that you, as a person thinking about how your company can become more self driving, can take away from this post.

首先，你需要重新思考AI能做什么——别再只把AI看作能改变个人或小团队工作效率的工具，而要承认，这项技术真正的影响力在于彻底重塑工作的组织方式。

The first part, I think, is the way that you think about what AI can do in general and to move out of thinking about AI as something that can change how individuals or even small teams work and instead assume that the true implications of this technology are huge structural change to how work gets done.

说起来容易，但在实际中很难做到。

This is easy to say, but hard to do in practice.

我们几十年来习惯了按固定模式做事，习惯了一点一点地改进。

We are wired by decades of doing things in certain ways to evolve only incrementally.

但这里的变革不是渐进式的改进。

But the implications here are not incremental evolutions.

而是从根本上改变了工作的本质，以及我们与工作之间的关系。

They are massive shifts in the very nature of what it means to work and what our relationship to work is.

在某种程度上，我们必须始终把这当作目标和方向，而不是让人害怕的东西。

And in some ways, we have to continually hold that as the implications and goal rather than something to be frankly scared by.

接下来，一个很自然的问题是：这难道只适用于天生就懂技术的公司吗？

Next, I think one of the most reasonable questions to come out of this is, is this for engineering native organizations only?

阿姆贾德提到的很多内容，都跟软件开发的具体工作紧密相关。

So much of what Amjad talked about is specific to the work of software development.

考虑到Replit不仅需要软件开发，还向所有人提供软件开发服务，这个问题就更值得深思了：这种模式是不是真的只适合技术主导的公司？

And given that Replit's business not only requires software development, but is in fact providing software development to everyone, there's even more of a reasonable question here, I think, about whether this is limited to engineering led organizations, let's say.

我稍后会再回过头来谈这个问题，但我想强调一点：不管怎样，Replit明确的做法是先从工程师入手。

I wanna come back to that question in a minute, but I will say that no matter what, one takeaway, which was explicit at Replit, is to start with engineers.

这不只是因为工程师对这项技术更熟悉，更因为他们的工作领域本身就非常清晰，而其他类型的工作则很难做到这一点。

And that is not only because they're more comfortable with this technology than others, but because the domain in which they work has clear around it in a way that some other categories of work is going to have a harder time with.

这一点我觉得挺直观的，但说得更具体一点：当你发现营销活动出了问题，问题往往很抽象。

This, I think, is fairly intuitive, but to put a fine point on it: When something is wrong with your marketing campaign, it very well might be intangible.

比如某个词或表达方式让你感觉不对劲，跟品牌一直想塑造的形象不符——作为市场人员或品牌守护者，你一眼就能察觉，但AI却很难自己判断出这里有问题。

A weird feeling about how a certain word or phrase hits doesn't really comport with the brand identity that you've been working to build, which you know as a marketer and a brand guardian upon seeing it, but which wouldn't necessarily be easy for an AI to pick up itself that there was a problem.

再看看软件漏洞，代码运行时会明确地出错。

Compare that to a software bug, where when the code runs, it does something verifiably wrong.

这种漏洞，AI修复起来要比那种模糊的营销问题容易多了。

That bug is gonna be much easier for AI to fix than the much more intangible marketing issue.

虽然我不认为这种自动驾驶式的工作方式应该只限于工程团队，但毫无疑问，最合适的起点就是工程师群体，因为那里最天然适合这种实验。

And while I don't believe that that means that this sort of self driving should be engineering only, it certainly does make sense to start experimenting with this process in the place where it is most natively fertile, which is going to be with engineers.

不管从哪里开始，阿布贾德几乎一带而过的一个关键点是：这一切的根本前提，不只是拥有一个能替你做事的智能体，而是要让这个智能体完全融入你现有的所有系统。

Now, wherever you start, one of the most important pillars that Abjad almost breezes past too quickly is the fact that the fundamental prerequisite for all of this is not just having an agent that can do stuff for you it's having that agent integrated with all of your existing systems.

这远远不止是能访问结构化数据那么简单。

This goes way beyond having access to good structured data.

而是需要全公司范围内的工具与系统深度整合。

It is full on cross org context tool and system integration.

阿布贾德提到，工程师能接入 GitHub、GCP、Azure、Linear、Notion、Slack、Zendesk 等各种工具。

Amjad talks about engineering giving access to GitHub, GCP, Azure, Linear, Notion, Slack, Zendesk, and more.

如果智能体连公司核心系统都访问不了，就根本谈不上帮助公司实现自动驾驶。

Without access to the systems that drive the company, there is no way for an agent to help the company become self driving.

就这样。

Period.

到此为止。

Full stop.

在该系统访问的上下游，还会出现一些安全问题，这些问题本身也需要解决，但系统访问必须被视为前提条件。

Now both upstream and downstream of that system's access are going to be things like security issues that need to be solved themselves, but that system's access has to be assumed as a prerequisite.

关于上下游的问题，下一个要点是，即使整个系统带来了新的机会，新的问题依然会出现。

Now, of upstream and downstream problems, the next takeaway, I think, is that new problems will emerge, even if overall, the system is unlocking new opportunities.

以Replicate的研究为例，当你生成大量新代码时，就需要审查大量新代码。

To take an example from the Replicate study, when you're producing new code, that's a whole lot of new code to review.

他们还建立了一套新的审查流程，同样整合了智能代理，不仅解决了问题，还提升了代码审查效率。

Now, talked about how they built a new review process that was agent integrated as well, solving the issue and in fact improving PR review time.

但这并不意味着这不是一个需要解决的新问题。

But that doesn't mean that that wasn't a new problem to be solved.

只是他们已经解决了它。

It just means they solved it.

我认为，很多试图转向这种自动驾驶模式的组织，会在意识到自己并没有摆脱问题时停步不前。

I think many organizations that try to move towards this self driving modality will get stopped when they realize that by moving to this agentic way of working, they're not giving up on problems.

他们只是换了一组问题来应对。

They're switching the problems that they have to deal with.

当然，这其中的权衡在于，希望这些新问题一旦解决，要么本身就是更值得处理的问题，要么就是解决了之后能带来一种明显更高效的工作方式，值得付出这个代价。

Now, of course, the trade off is that hopefully those new problems once solved are either a, better problems to be solving or b, problems that once solved are in service of a better way of working that is so clearly and self evidently better that it's worth doing.

但自动驾驶并不意味着没有问题。

But self driving does not mean no problems.

它意味着会出现新的问题。

It means new problems.

明白了。

Okay.

所以你现在已经假定了一种巨大的结构性变革。

So you've now assumed huge structural change.

你从工程师开始入手。

You've started with your engineers.

你已经给系统开放了权限，也接受了会冒出新问题并需要解决的事实。

You've given systems access, and you've accepted that there will be new problems that emerge that need to be solved.

那接下来，你怎么把这种做法从工程团队推广到其他部门呢？

How do you then move from the engineering section of their organization to others?

这篇帖子里特别值得注意的一点是，他们的做法明显是吸引人主动参与，而不是强行推动。

Another thing that stood out from this post is that their approach was clearly pull, not push.

换句话说，他们通过像 Slack 这样的公开平台，让工程师和新系统互动，其他部门的人就能自然看到这些互动带来的好处，而不需要先被要求全身心投入才能被说服。

In other words, by using public spaces like Slack for engineers to interact with the new self driving system, they allowed others from other parts of the organization to see the benefits of those interactions without, it seems at least, demanding that they dive all the way in before being convinced.

我觉得我们在面对重大结构性变革时，常常低估了‘用行动展示’比‘用语言说服’的价值。

I think we sometimes underestimate how valuable it is to show, not tell when it comes to the benefits of new, big structural changes.

人们天生就容易抗拒改变，尤其是工程部门以外的人，虽然也不仅限于他们。

People inherently, especially outside of engineering but not exclusively, are going to be, in many cases, reluctant to change.

这既是人性使然，也是大企业固有的惯性。

That is just human nature and it's certainly enterprise inertia.

但只要创造一些空间，让大家不仅能看见新工作方式带来的成果，还能亲眼看到工作是怎么实际推进的，很多疑虑就能提前化解。

But by creating spaces where people can see not only the results of that new way of working, but how the work actually proceeds itself, a lot of that skepticism can be dealt with in advance.

一旦工程部门以外的人开始感兴趣，他们也得像工程师一样，把各自的系统连起来。

Now once people outside of engineering do start to get interested, they, like engineering, are also going to have to have their systems wired together.

在某些情况下，这会是一个技术性的工作。

And in some cases, this is going to be a technical process.

工程师可能觉得给这些新智能系统接入正确的 API 是理所当然的，但对其他团队来说，这可能完全不直观，甚至技术上都难以实现。

While it might be second nature for engineers to be able to give these new agentic systems access to the right APIs, that might not at all be intuitive or even technically feasible for these other teams.

我觉得一个越来越重要且有价值的做法，是让工程师——尤其是那些已经经历过这种转变的工程师——与需要帮助的业务部门配对，协助他们建立新的工作方式。

One pattern that I think is going to be increasingly important and valuable is the pairing of engineers, especially engineers who have already gone through this sort of change, with parts of the business organization that can use their help to set up the new way of working.

我觉得，比起外派的工程师，内部派驻的工程师在推动这种全面的结构性变革时，同样重要，甚至可能更重要。

I think as important as forward deployed engineers are, internally deployed engineers are going to be every bit, if not more important, when it comes to this full scale structural change that's happening.

但一旦整合了所有上下文信息的系统和数据源上线，这些非工程团队依然需要支持，来设计真正能实现自动驾驶的反馈闭环。

But once the systems and data sources that have all the context come online, those non engineering teams are still gonna need support designing the types of loops that can actually result in true self driving.

我真的很喜欢这篇博客文章。

Now, I love this blog post.

我觉得写得特别好。

I think it's great.

但如果非要说哪里还能再深入一点，我觉得是把这个自动驾驶的类比进一步深化。

But if there's a place that I would like to see it push even farther, it's actually deepening this analogy of self driving.

阿姆贾德提到，智能代理要承担大量新任务。

Amjad is talking about agents taking on a lot of new work.

但自动驾驶不仅仅是代理在干活。

But self driving isn't just about agents doing work.

真正的关键是：代理的目标由人设定，能访问完成目标所需的系统和数据，并且有明确的评估标准，来判断自己做的事到底有没有达成目标。

It's about a pattern in which agents have their goals set by people, have access to the systems and data they need to do the things that are required to achieve those goals, and have the criteria to see how what they're doing is actually achieving those goals or not.

我们说的‘闭环’，指的就是这个。

Now, this is what we talk about when we talk about loops.

闭环之所以不只是个流行词，是因为它本质上是一种结构化目标并设置可验证结果的互动模式，让代理能更自主地完成更多工作。

But the reason that loops aren't just a buzzword is that more than anything else, they're this interaction pattern of structuring goals with verifiable endpoints so that agents can do more work on their own.

而当这些闭环能够接入持续更新的客户信息流，让目标实时调整时，它们才真正从团队层面的流程，上升为整个自动驾驶型公司的底层机制。

And where I think loops jump from just being an individual or team process to actually being the underpinning of a self driving company is when they're given the access to the ongoing continuously updated flow of customer information that allows the goals of the loops to evolve in real time.

这就是为什么阿姆贾德说，他们AI团队做的那个持续学习系统最让人兴奋——它能分析用户反馈、提出改进方案，并通过基准测试和A/B测试来验证效果。

This is why Amjad said that the most exciting self driving example came from their AI team, who built a continual learning system to analyze user feedback, propose improvements, and use a combination of benchmarks and AB tests to validate the wins.

我认为，这篇文章里最能体现我心目中全自动驾驶理念的例子，就是这个了。

That, I think, is the best example of full self driving as I imagine it from the entire article.

除了这种全自动模式之外，很明显，这样做还有一个好处，就是让大家意识到，每个团队的工作在不同程度上都在拖其他团队的后腿。

Now, outside of that full self driving mode, it's clear that one of the other benefits to doing this is the recognition of just how much every team's work is to greater or lesser extent being a bottleneck for other teams.

最典型的例子就是，所有非工程团队都得依赖工程团队来开发东西，但实际上，每个团队都会被其他团队找上门，要求提供信息、更新数据，或者对齐目标。

Now, the classic example of this is all the non engineering teams needing engineering to build something, but every team gets called on by every other team to ask things or get access to updated information or get aligned with goals.

我觉得，这种全自动理念的一大价值，就在于能让团队不再互相卡脖子——把大量日常流程交给智能代理来处理。

And I think a big value of this sort of self driving idea is to allow teams to stop being bottlenecks for one another by having a lot of that operational flow run through agents instead.

不过，我们还得再提一次，Replit 在做这些实验方面确实有独特优势。

Now it is worth noting once again that Replit is in a unique position to be doing a lot of these experiments.

他们不仅是家软件公司，还是专门开发软件开发工具的软件公司。

Again, not only are they a software company, they're a software company that builds tools for building software.

他们就像站在风暴的正中心，没人比他们更贴近这场变革的核心。

They are in the eye of this storm like no one else.

但他们为自己打造的很多解决方案，最终都会变成产品推向市场。

But a lot of the solutions that they built for themselves are going to be productized.

顺便说一句，我猜Replit自己也会把其中很多东西做成产品。

And by the way, I assume that a lot of them are gonna be productized by Replit themselves.

但即便如此，我还是能想象很多高管读了这篇文章后会想：我们可没有那种能开发出和SaaS竞争的工具的工程团队，但我认为你其实不需要有。

But still, I can see a lot of executives reading this and thinking to themselves, well, we don't have the engineering organization that can actually go build tools that could compete with SaaS, but I don't think you're going to need to.

我认为，哪怕你觉得工程资源不够，也千万别忽视这种‘自动驾驶’的思维方式，那会是个错误。

And I think it would be a mistake to not at least explore thinking in this self driving way because you don't think you have the engineering resources or wherewithal to do so.

我觉得，如果我们发现这种公司层面的‘自动驾驶’模式真能在各种类型的企业中带来这些收益，那市场就会有巨大的动力去把这套流程的各个部分产品化，所以可以肯定，未来六到十二个月内，所有这些组件都会以某种形式出现在某个工具里。

I think that to the extent that we find that the self driving mode for companies actually produces these sort of gains across lots of different types of companies, that there is such an incredible market incentive to productize parts of that process that you have to assume that over the next, honestly, six to twelve months, all the pieces will find their way into some tool somewhere.

换句话说，只要你开始往这个方向走，哪怕你的工程团队能力有限，也总会有人帮你解决这些具体问题。

In other words, I think you can assume that if you start to wander down this road, even if you have severe limitations in your engineering organization, there will be answers to those particular problems.

我们一开始谈的是一个思维上的启发。

Now, we started with a mindset takeaway.

现在，我们最后再回到一个思维上的启发。

We're going to end with a mindset takeaway.

归根结底，这一切都是在改变我们对‘什么是可能的’这种固有认知。

I think ultimately this is all about changing our assumptions about what is possible.

这些变化初期可能表现为代码量增加、新功能上线时间缩短、客户支持问题解决率提升等等。

A lot of these things will look like increased lines of code and decreased timelines for new features, improved resolution rates on customer support, and all of that at least at the beginning.

但如果我们真能建立起闭环系统，不仅能连接公司内部，还能对接外部环境，那么几年后，企业进化和改进的速度将与今天截然不同。

But if we can really set up looping systems especially that interact with not only the inside of the company, but outside the company as well the rate at which companies can evolve and improve will be totally different a few years from now than it is today.

这当然会带来各种新挑战，但跟其中的机会相比，这些困难都微不足道。

That is, yes, going to create all sorts of new challenges, but they pale in comparison to the opportunity.

如果你现在正在听这段内容，那你正处在绝佳的位置，有机会成为率先利用这一趋势的少数人、领导者或企业之一。

And I'll tell you what, if you are listening to this right now, you are in a unique position to be one of the people and one of the leaders and one of the companies that takes advantage of that before everyone else.

所以，在结束之前，去让你的智能助手安排一次会议，跟公司所有高管讨论一下‘自动驾驶’对你公司意味着什么，然后告诉我你们的发现，我可以为此做一期后续节目。

So now as we wrap up, go ask your agent to schedule a meeting with all of your top execs about what self driving means in the context of your company, and then tell me what you figure out so I could do a follow-up to this episode.

好了，今天这期《AI每日简报》就到这里。

For now, that's gonna do it for today's AI Daily Brief.

一如既往，感谢你的收听或观看。

Appreciate you listening or watching as always.

我们下期再见。

Until next time.

再见。

Peace.