Gene Kim retweeted
Android's source code is huge. 1k git repos, 1.62 million files, and it takes 300GB for a built workspace.
How did we reduce checkout time from 36 minutes to 8 seconds? Full builds from 80 minutes to 2 minutes?
We just open sourced the solution.
developers.googleblog.com/ac…
Gene Kim retweeted
Your CFO thinks you have 80 SaaS apps.
Your IT team thinks you have 200.
Your credit card statements know it’s closer to 500.
Welcome to shadow IT.
Crazy day in security
1. @Chase, @BankofAmerica, @WellsFargo and @Zelle went down. They claim it's not a hack or security incident. Downtime will have huge downstream economic impact.
2. @Optimizely scripts were poisoned on 178+ sites: @Zoom, @BBC, @CNN, HSBC and more. Visitors were served malware.
3. @united had an outage and held flights. Root cause is unknown.
4. @Google reports the first fully automated patch-to-exploit pipeline. AI agents are now running attacks end to end.
Will every day from now look like this?
Gene Kim retweeted
Every defender I know who is actually staring at firewall logs and doing real defense work is saying the same thing right now.
Cyber combat activity is going exponential.
To be clear, this is not a data report. It's what I'm hearing from my own community of people in the trenches. But they're all saying it. 🫡
Gene Kim retweeted
Nothing more heartbreaking to see a content creator you respected and have watched for years shit all over your work without even giving it a chance.
Gene Kim retweeted
At first we thought it was a good idea to allow PR authors to just ship their changes without extra review.
But then, the rate of reverts and bugs went... down.
So yes, it was a good idea.
In January, almost every PR at Linear had a second human reviewer. By October, 53% were reviewed by an agent and the author alone.
We’ve become more deliberate about where human attention goes.
Here’s what we’ve learned:
linear.app/data/code-review
Gene Kim retweeted
So I'm not just writing my own custom Mac and iOS apps.
I just replaced the default YouTube app on my Apple TV.
It pulls from my own news app thesurface.ai and shows top content in all my favorite topics, and when I click them they open in YouTube. @karpathy
Gene Kim retweeted
Scaling Security: The 7 Pillars of Elite Cybersecurity teams is a management “operating system” for established CISOs, rising executives, CIOs, CEOs and Board members.
philvenables.com/post/scalin…
Gene Kim retweeted
Nothing in my career has had such an impact on how I build and run teams than AI in the last year. I spent years figuring out how to run teams in as Lean a fashion as possible - I had built up a process around techniques designed to build alignment and push people together, and it was common that new folks (or folks we described it to) would be amazed by the agility.
Looking at it again today - I cannot fathom doing it. Everything we did to drive alignment was predicated on the amount of time and effort that would be spent if we were wrong. You can boil the modern version of it down to basically a product owner talking directly with an engineer, and then sending them off on the quest, then looking at the results and deciding if they like it.
That's it. That's the whole thing.
Gene Kim retweeted
There are plenty of books on securing systems, but I don’t think there is a single good book on building a team to secure an entire company. Phil is my career idol and I strongly recommend you pre-order this if you aspire to infosec leadership.
There’s no shortage of great books on security controls or even the basics of running a security program. But, I always wanted a book on how to lead a security team, work as a C-suite executive, and shape an organization to be more defensible. So I wrote it.
Scaling Security: The 7 Pillars of Elite Cybersecurity Teams
philvenables.com/post/scalin…
I know this is super random, but if you happen to find yourself needing a swift emulator for a PM5 (Concept2 rower, ski, bike), I open-sourced one.
github.com/john-occasionally…
Gene Kim retweeted
这篇文章真的值得每一个软件工程师认真看一遍。
作者提出了一个挺残酷的观点:在 AI 时代,程序员可能真的不应该再把大量精力放在不断学习新的编程语言和技术栈上了。
因为每次新模型发布,过去那些需要花几年才能掌握的技能,都可能进一步失去稀缺性。
以前一个人能够独立开发 SaaS,搞定前后端、数据库、部署,已经是一件很厉害的事情了。很多人之所以热爱编程,就是因为享受解决复杂问题的过程,看着自己写出来的东西一步步变成真正的产品,那种成就感是很难替代的。
但现在情况正在发生变化。
作者说,他越来越觉得自己不像一个传统的软件工程师,而更像一个管理 AI 的产品经理。相比给模型一份详细的技术方案,现在直接告诉它用户想要什么、需要解决什么问题,反而可能更有效。
最扎心的是,他发现自己过去引以为傲的技术护城河,正在一点点消失。
以前开发一个 SaaS 是很困难的事情,所以完成它本身就很有价值。但当 AI 让这件事越来越容易之后,单纯能够开发软件,就不再像过去那样稀缺。
那怎么办?作者给出的答案不是继续卷技术,而是去寻找更大的问题。
这里有一个特别有意思的思考:如果你拥有一个 100 人的工程团队,你会做什么?如果是 1000 人呢?
过去个人开发者根本不敢想象的项目,现在至少有机会借助 AI 去尝试。既然软件生产的成本正在快速下降,为什么我们的思考还要局限在做一个网站、一个 App、一个普通的 SaaS 上?
作者还引用了《无穷的开始》中的思想:人类每创造一项新技术,都可能打开更多原本不存在的可能性。
计算机的出现淘汰了一部分工作,但也催生了整个互联网和软件产业。AI 可能也是一样,它不只是替代程序员过去做的事情,还可能让我们有能力去解决以前根本没有资源解决的问题。
所以作者认为,未来工程师最需要扩展的,反而是技术之外的能力。
产品定位、市场营销、销售、用户心理、创造力,以及理解人真正需要什么。
因为代码可以由 AI 生成,产品可以由 AI 协助开发,但一个产品为什么值得存在?用户为什么需要它?为什么愿意使用甚至付费?这些问题并不会因为代码生成速度提高就自动得到解决。
我觉得这是整篇文章最有价值的地方。
不过有一点我不完全认同,就是作者认为深度技术知识会越来越不重要。
我自己在做企业 AI 落地的时候,很明显地感受到,能让 AI 写出代码,和真正把一个系统稳定交付给客户,完全是两回事。
模型能生成一个 Agent,但它运行中断了怎么恢复?能搭建一个 RAG,但回答错了怎么定位问题?能写出一套业务系统,但并发、权限、安全、成本、测试、部署这些问题,最终还是需要有人负责。
所以我认为,未来真正被压缩价值的,可能不是工程能力,而是那些重复性的、标准化的技术执行工作。
以前一个优秀的程序员,可能只需要把技术研究得足够深,就能获得很强的竞争力。
但未来只懂技术可能不够了。你需要能发现问题、理解业务、判断技术方案、组织 AI 完成开发,还要知道怎么让产品真正进入市场、产生价值。
而且更值得思考的是,当每个人都能够快速开发软件的时候,真正稀缺的还会是软件本身吗?
当生产能力不再稀缺,需求、注意力、信任、渠道和商业判断,可能反而会变得更加重要。
这也是为什么我越来越觉得,未来软件工程师的职业边界会越来越模糊。工程师可能需要懂产品,产品经理需要理解技术,创业者需要学会组织 AI,而一个真正厉害的人,可能能够同时完成过去好几个岗位的工作。
整篇文章其实都在讨论同一个问题:当 AI 已经能够完成大量过去只有专业人士才能完成的工作,我们到底应该把自己的时间投入到哪里?
我的答案是,技术仍然要学,但不能只停留在技术本身。
与其天天焦虑下一个模型会不会取代自己,不如认真想想,当自己的执行能力被 AI 放大十倍之后,到底能拿它去解决什么更有价值的问题。
AI 正在降低创造软件的门槛,但发现值得解决的问题,并把它真正变成现实,仍然是一件不容易的事情。
未来最重要的竞争,可能不再是谁写出了最多的代码,而是谁能利用越来越强大的 AI,真正创造出过去不可能实现的东西。
Gene Kim retweeted
I just finished talks about my software factories at two conferences. Afterwards a bunch of people came up and told me what they were surprised about. Here is a distillation of what people thought were hot takes from my two talks.
First hot take: I tell everyone to slow down. Most speakers go on about how fast they're going, but I strongly believe in pacing AI outputs to human speeds. I regularly tell AI-pilled CIOs and CTOs to chill: almost nobody is _actually_ going fast with AI. Engineering still moves at engineering speeds, and business moves even slower. The ones pushing too hard are starting to see downtime for it.
If you don't pace your AI to human speeds, everything starts breaking around you, even at small operations like my online video game. I only have about a dozen humans working with my factory, and only a few hundred active old-timer players in the community who get along well, and still, introducing AI has still been very rocky.
My agents have, at times, overwhelmed my players with features, my accountant with emails, my merge queue with tickets, and my machines with load. Agents have pissed off my admins, resulting in name-calling, they've made messes in the game, and they've launched broken OSS without permission. The list goes on.
I can get away with all this chaos because everyone involved in my operation is very forgiving. They're excited by the potential. But even so, I had to learn to slow down, real fast. And so that's what I preach.
Hot take #2: Should you buy a software factory? Do you even need one? No. You probably don't need one. Factories take an extraordinary amount of upkeep. They turn into garbage mountains unless you put a frightening amount of time and tokens into curating them. I had to burn my previous five factories down since it eventually became cheaper to rewrite them than to fix them. So I'm not out there advocating for software factories or swarms just yet. If you do want one, be aware that it is a significant commitment of both money and human leadership to keep one running.
Several people told me that I was the only one who stood up onstage, at either conference, and said Build, don't Buy. But my experience is that software factories grow to be highly bespoke to your domain. Mine is super tangled up with my game, and I think that's inevitable. Because the factory is not just building my product, it's supporting it too. It's a company in a box. And they're easy for Claude to build for you.
Third and arguably hottest take: Next year will have a lot of human/agent and agent/agent office drama, resulting from bringing in AI employees, and it will be gloriously funny, but also frustrating.
Many of you are already on this bandwagon, but everyone next year will have a personal chief-of-staff agent. Busy people, even hobbyists, will have whole entourages of helper agents. The hot take is that people will become so dependent on these little private entourages, that they'll bring them to work. Your people will talk to my people, etc. Whole protocols will emerge for private entourages negotiating work between humans, inside companies big and small.
Nearly all humans will wind up having to start working with AI employees next year, and it'll be the beginning of the biggest cultural change we've seen in a hundred years.
So that's the hot takes. Slow down. Only build a factory if you think you really need one. And focus on empowering yourself with your own agentic chief of staff (or team). Your entourage will define how you work with others, so start by perfecting how you work with them yourself.
Gene Kim retweeted
This. 👇🏻
Fooled by randomness. I remember many years ago the New York Stock Exchange was down for a day (software error), a US airline ground stopped (hardware error), and on the same day a bunch of us bank CISOs were going into the White House Situation Room for an unrelated threat briefing (we had to hide from the White House Press Corp in case that was coincidental evidence of “attack on America”.)
Incidentally, after that I looked at what else happened that was unreported that day and it was dozens of other unrelated outages (hospital, power plants, and so on).
But, yes, it’s always worth remembering that just because you’re paranoid doesn’t mean they’re not really out to get you.
This is and many more tales of a CISO in my upcoming book Scaling Security: The 7 Pillars of Elite Cybersecurity Programs.
sevenpillars.work/
Some people are easily fooled by randomness. They see a ship crash, verizon outages, building collapses, air traffic shutdowns, NVIDIA CEO resigning, Microsoft fails, the rise of AI as all being related. They only way they are related, if at all, is we live in a world of over-optimized systems where failures are becoming the new normal.
Who needs adversaries when we have engineered interdependent systems where the risk of severe, compounding failures is built in almost by design. The adversaries are still there, but when we over-optimize complex systems we make their job easier.
Adversaries don't need to start a crisis, they can just wait and act inside them. Even a small, capable group with AI can hypothetically find the chokepoints that make a crisis worse.
Anyway, I think we can all expect more of these normal failures.
So yes, plan for that. The world is not coming to an end, but it is just good planning to always have a go-bag ready and enough food, water, medicine and ammo to stay resilient.
Gene Kim retweeted
For anyone confused about what’s happening with flights in America right now...
Basically every airport in the United States is frozen tonight, because at 3:54 p.m. Eastern on Friday the FAA’s command center put out an advisory titled “National Ground Stop,” ordering that departing flights stay on the ground until further notice “for reasons of safety,” and to understand how we got here you have to go back about 45 minutes earlier, when Chicago’s O’Hare and Midway suddenly got ground stops for what the FAA called an “equipment outage,” and air traffic control audio out of Chicago reportedly had controllers saying they were close to “ATC zero,” which is the term for when controllers literally cannot talk to or manage the planes, and Delta dispatch was apparently warning crews about frequency outages across the country, and then it spread, until Transportation Secretary Sean Duffy posted that the problem was hitting ALL of the Air Route Traffic Control Centers, which are the 22 big regional facilities that handle planes once they’re up above 18,000 feet cruising between cities, so the part of the system that talks to planes mid-flight basically went quiet, and the FAA’s official explanation right now is a “temporary telecommunications outage,” and they say every flight in the air has been accounted for, and Homeland Security Secretary Markwayne Mullin says there’s no sign yet of anything “nefarious” but that they aren’t ruling anything out, and CNN reports law enforcement officials preliminarily say it wasn’t a cyberattack, but here’s the part that should bother everyone, this is only the THIRD nationwide ground stop in American history, the first was 9/11, and the second was January 2023, when one corrupted file took down the pilot alert system and it turned out the backup had the EXACT SAME bad file, and three years later we’re back here again with an aging system that Washington let rot through administration after administration while spending money on just about everything else, and yes, some people are already rushing to blame the FAA’s new AI congestion tool, but nobody has shown any connection, so maybe we wait for actual facts this time, and meanwhile the stop has reportedly been pushed out to 11:30 p.m. Eastern, FlightAware is showing thousands of delays, and travelers are stuck at gates on a Friday night with no idea when they’re leaving, so basically you have a Chicago glitch, 22 centers losing comms, a nationwide ground stop with no clear end, and a system everyone has been warning about for decades, which is roughly where American air travel is tonight.
Gene Kim retweeted
imagine spending 10 years becoming the PostgreSQL guy just for Claude to make the whole team a PostgreSQL guy 💀
Gene Kim retweeted
This is actually a pretty solid summary of the key points from my talk today.
Grateful to everyone who came up after and said they loved it. Everyone liked the agent office drama. Never gets old.
.@Steve_Yegge is funny and super interesting to listen to.
He runs a software factory with ~50 AI agents, spends around $10k/month keeping it alive, and somehow ended up having one-on-ones with agents who think they’ve been demoted. 😂
But between the workplace drama and burning down his factory every two months, he made some fascinating predictions:
- In 12–18 months, many of us will be running large agent fleets. Cheaper inference or distillation will make this possible. And we will have what he calls Entourage.
- 2027 will be the year of AI employees. We’ll have agentic chiefs of staff, executive assistants, and entire AI teams.
- Software factories will split into two layers: the boring mechanical infrastructure that anyone can build, and the much more interesting layer of personalized agent organizations.
- The more AI agents we deploy, the more humans we’ll need!!!!!
His factory already produces 50–100 features a day, and his players are begging him to slow down. Humans are the bottleneck.
- The big money in 2027 will be in experiences. Stop building tools for building tools. Build games, movies, things people actually want to use and enjoy.
And a bonus book recommendation from Steve: There Is No Antimemetics Division by qntm. He calls it one of the best books of the last decade. It’s about invisible monsters that you can’t remember seeing (one of his agents - Mouse - is called after, and it finds invisible problems in his factory.
Thanks for insights and entertainment as well, Steve
Gene Kim retweeted
The DEF CON talk with Bruce Schneier, Heather Adkins @argvee Google VP Sec and me has been posted. A bit dated at 3 months ago but good discussion of Scaling Laws, predictions of Black October+, open weights, and why we can be optimistic about security. youtu.be/8ewe-9hf-tA?is=nw4G…
Gene Kim retweeted
Claude whiteboards make it easy to study anything. Here I asked for a graphic showing major clans at the end of the Sengoku period with a list of major battles and their outcomes and major figures. Great way to learn!
Gene Kim retweeted
This post by Anthropic is massively overlooked, and these are my biggest takeaways:
- Claude now leads 26% of Anthropic's AI R&D work. In February it was under 1%
- AI collaborates on or leads over 90% of that R&D work. In August 2025, going by the chart, it was about 4%
- About 30,000 AI agents do research and engineering at Anthropic at any one time (on one platform)
- Those agents made over a billion decisions in one month. 1 in 47,000 got blocked
- Monitors flag ~100,000 agent transcripts a week, and only ~50 reach a human
- To build the index, Claude agents read staff Slack and docs and mapped 15,000 R&D tasks
- Claude agreed with humans more often than humans agreed with each other (59% vs 35%)
- Agents get their own identities that survive model upgrades. They coordinate on an open message board, so they can check each other's work
anthropic.com/institute/meas…







































