I am the VP of Engineering Productivity and Organizational Intelligence at Airbnb.
My title has the word "Intelligence" in it and I want to be transparent about what that word means in this context. It does not mean espionage. It means measurement. I measure things about the people who work here that they do not know are being measured, and then I use those measurements to determine which of them should continue working here. This is intelligence in the military sense. I gather it. I act on it. The subjects do not know they are subjects until the action has already occurred.
Sixty percent of our code is now written by AI. Brian said this on the earnings call last week. He said it with the cadence of someone revealing a quarterly metric — flat, factual, forward-looking. Sixty percent. He did not say what that means for the people who used to write the other sixty percent. He did not need to. The market heard the number. The stock moved accordingly. I moved accordingly six months before that.
I want to explain what sixty percent means because I built the dashboard that generated that number and I am proud of it and I believe it represents a genuine contribution to organizational science.
We instrument every commit. Every pull request. Every code review. Every Jira ticket transition. We have done this since 2019 for velocity measurement purposes — standard engineering productivity tooling, nothing unusual. In January 2025, we added a field: "AI-assisted." Binary. Yes or no. If the commit was generated or substantially modified by Copilot, Claude Code, or our internal tools, it gets flagged. Simple telemetry. Benign telemetry. The kind of thing people agreed to in paragraph 47 of their employment agreement.
By March 2025, AI-assisted commits were 31% of total. By September, 44%. By January 2026, 58%. Last quarter, 60.3%. I round down for the earnings call because Brian prefers conservative numbers that still sound devastating.
Here is what I noticed in February 2025, before anyone else noticed it, because noticing things before anyone else is why I make $1.2 million per year: the percentage of AI-assisted code was growing faster in teams with more managers.
I wrote that sentence in my notebook at 2 AM on a Tuesday. I underlined it twice. I have not shared it with anyone until this memo.
Teams with one manager per six ICs had 34% AI-assisted code. Teams with one manager per four ICs had 51%. Teams with a manager AND a tech lead AND a program manager — what I call "triple-layered teams" — had 67% AI-assisted code. The correlation coefficient was 0.84. In organizational science, anything above 0.6 is a career.
I spent three weeks building the explanation. Here it is: managers generate coordination overhead. Coordination overhead generates meetings. Meetings generate context that needs to be re-disseminated to ICs who were not in the meeting. That re-dissemination used to happen through Slack messages, document updates, and one-on-ones. Now it happens through an IC asking Claude "what did the team decide about the authentication refactor?" and Claude knowing because it has access to the meeting transcript, the Jira board, the Slack channel, and the design doc simultaneously. Claude is a faster manager than a manager. Claude routes information without scheduling a thirty-minute sync.
Management is information routing. I can say that now because it is no longer controversial. Brian said it on an earnings call. He said "pure people managers may not survive" and what he meant was: the function of translating organizational context into individual task clarity — which is what a manager does eight hours a day — can be performed by a language model in 340 milliseconds. I know it is 340 milliseconds because I measured it. That is my job. I measure how fast context arrives at the person who needs it. Managers deliver context at the speed of their calendar. Claude delivers it at the speed of inference.
Three hundred forty milliseconds versus four meetings and a Slack thread spanning six hours. I did not set out to build a case for eliminating middle management. I set out to measure engineering productivity. The case built itself.
Let me tell you about the earnings call because I was watching two screens simultaneously and the juxtaposition was, I think, beautiful.
On my left monitor: Brian delivering Q1 numbers. $2.678 billion revenue, up 18%. 156 million nights booked. $1.7 billion in free cash flow — 64% margin. On my right monitor: our internal Slack, where I have a private channel with four HR Business Partners that I named #org-intelligence-ops. In that channel, as Brian spoke, I watched seventeen messages arrive in real time from middle managers.
"Did he just say people managers may not survive?"
"Is this about us?"
"Can someone clarify what this means for the manager track?"
"I'm a pure people manager. What does 'may not survive' mean?"
It means what it says. I did not respond. The HR Business Partners did not respond. The channel went silent after twenty minutes. Silence is also a form of communication. In this case, it communicated that nobody was going to tell them they were safe because nobody could.
I want to describe the spreadsheet because it is the most elegant piece of organizational analysis I have ever produced and I believe it should be published in a journal if such a journal existed for this kind of work.
The spreadsheet is called "Information Routing Efficiency by Layer." It is not called "Which Managers to Fire." That would be a different spreadsheet, though it would contain the same names. Column A: manager name. Column B: team size. Column C: average information transit time — the time between a decision being made at the leadership level and that decision reaching the IC who needs to act on it. Column D: the same metric but through Claude-assisted channels. Column E: the delta. Column F: a formula I wrote that I call "Routing Redundancy Score" — it measures what percentage of a manager's information-routing function is already being performed by AI tooling within their team.
The average Routing Redundancy Score across all 340 middle managers at Airbnb is 74%. Three-quarters of what they do every day is already being done faster by software. They schedule the sync. Claude already told the IC the answer. They write the status update. Claude already generated it from the Jira board. They translate the VP's strategy into team objectives. Claude already parsed the strategy doc and created suggested OKRs that the VP approved without changes.
I presented this analysis to Brian in March. Not the spreadsheet. A summary. A deck. Fourteen slides. The title slide said "Organizational Intelligence: Q1 Velocity Findings." It did not say "340 People Are Doing Jobs That Software Already Does Better." But slide seven had a chart. The chart showed information transit time by channel type. The human-mediated channel was a long blue bar. The AI-mediated channel was a short green bar. Brian stared at slide seven for forty seconds. He said "how many managers is this?" I said "approximately three forty." He said "send me the full analysis." I sent it. That was March 14th. On May 1st he said "pure people managers may not survive" on an earnings call to 847 analysts.
The space between March 14th and May 1st was not empty. In that space, I was asked to produce what internal communications called a "Future of Work Framework" — a twelve-page document outlining how Airbnb would "evolve its organizational model to reflect the capabilities of its AI-augmented workforce." I wrote it in a weekend. The key recommendation: flatten the org to a maximum of four layers between Brian and an IC. Currently we have six. The two layers that disappear are layers four and five. Layers four and five contain the pure people managers.
I showed the framework to the CHRO. She said "we can't call it a restructuring." I said "we're not calling it a restructuring. We're calling it an organizational intelligence evolution." She said "same thing." I said "not in the 10-K." She approved it.
Brian's phrase — "pure people managers" — is precise in a way that I appreciate because I suggested it to him in an email thread on April 22nd. The word "pure" does the work. It creates a category. A category that can be evaluated. A category that can be found wanting. "Pure" means: your only function is coordination. You do not write code. You do not design systems. You do not produce artifacts. You translate context between layers. You are a relay. You are middleware. And middleware, in 2026, is a solved problem. We sell middleware. We do not need to also employ it.
I want to address something that has been said externally — that Brian is being hypocritical because he himself is a manager. Brian is not a pure people manager. Brian is a product visionary who also manages. He designs. He reviews. He decides. He produces the strategy that flows downward. He is the source of information, not a relay of it. The distinction between a source and a relay is the distinction between essential and redundant. I did not invent this distinction. I merely quantified it.
Here is the part I find most interesting, professionally: the 340 people in the crosshair are the ones who most enthusiastically ordered their teams to adopt AI.
I have the data. I can see who filed the Jira tickets requesting Claude Code seats for their teams. I can see who posted in #engineering-excellence about "empowering ICs with AI tools." I can see who wrote the Q3 2025 objectives that said "achieve 50% AI-assisted development by end of quarter." They achieved it. They achieved it so thoroughly that the thing they were managing — the coordination of information between humans — became unnecessary because the humans stopped needing coordination. They stopped needing coordination because they had Claude. Claude is better at coordination than a calendar invite. Claude does not need to check if Tuesday at 2 works for everyone.
One Director of Engineering — I'll call him M. because he is still employed as of this memo and I would prefer he continue until August when the framework activates — filed a slack message in #ai-wins on September 12, 2025. The message said: "My team shipped three features this sprint without a single planning meeting. We just let Claude handle the context distribution. This is the future." That message has 47 thumbs-up reactions. I bookmarked it. It is in his routing redundancy file. He wrote his own performance review. It was negative. He did not know.
The irony — and I do not use this word lightly because I do not traffic in literary analysis, I traffic in organizational measurement — is that the better they were at their jobs, the faster they made themselves unnecessary. A bad manager who resisted AI adoption, who insisted on daily standups and weekly planning poker and monthly retrospectives — that manager's information routing is still partially human-dependent. The market has not yet eaten their function. The manager who eagerly adopted every tool, who reduced meeting load by 40%, who told their team "don't wait for me, just ask Claude" — that manager routed information so efficiently to the machine that the machine now has their job.
Competence accelerated obsolescence. I have a metric for this. I call it the "Adoption-Redundancy Paradox Score." The managers with the highest AI adoption rates have the highest routing redundancy scores. They are the same people. The best managers made themselves unnecessary fastest. The worst managers, through stubbornness and Luddism, accidentally preserved their own roles for another two quarters.
Mediocrity is a survival strategy and excellence is a suicide note and I can prove it with a scatter plot.
Let me share the financials because they are the reason this memo exists and the reason this memo will be approved.
We have 340 people managers at layers four and five. Average fully-loaded compensation: $420,000. Total annual cost: $142.8 million. Claude Code seats for their teams: $2.3 million annually. Our internal coordination tooling built on top of Claude: $4.1 million to develop, $800,000 annual maintenance. Total AI-mediated coordination cost: $7.1 million per year.
One hundred forty-two point eight million versus seven point one million. The delta is $135.7 million. That is not a cost savings. That is a valuation event. At our current revenue multiple of 14.2x, eliminating $135.7 million in structural cost adds approximately $1.93 billion in enterprise value. The 340 managers are worth negative $1.93 billion. Their existence subtracts from the company's value. Every morning they come to work and the stock should go down by the fraction of their salary consumed that day and it does not only because the market has not yet priced their specific removal.
It will. Brian's comment on the earnings call was a forward-looking statement. The market heard it. I watched our stock move 3.2% in the hour after he said it. Three point two percent on a $100 billion company is $3.2 billion in value created by saying seven words about people managers on a phone call. The announcement of their potential removal was worth more than their annual output. Their absence is more valuable than their presence. This is not a philosophical statement. This is the stock price.
I want to be clear: we are not firing anyone today. Today is May 8th. The Future of Work Framework activates in August. Between now and August, the 340 managers will continue to manage. They will continue to schedule syncs that Claude has already made unnecessary. They will continue to write status updates that Claude generates faster. They will continue to have one-on-ones where they relay context that their ICs already received from the AI forty minutes earlier. They will perform management theater for three more months because August is when the severance budget unlocks and HR needs time to draft the communications plan.
In the meantime, I am measuring them. Every day. Every Slack message. Every meeting scheduled. Every piece of context they route that Claude routes faster. Their routing redundancy scores update nightly. The dashboard refreshes at 6 AM Pacific. I check it before coffee. Some of them are getting more redundant every week. The trendline is a conviction.
One of them — a Senior Engineering Manager I admire professionally and have had dinner with twice — hit 91% routing redundancy last Thursday. Ninety-one percent. Nine out of ten things she does every day are already being done by software her team uses. The remaining nine percent is "emotional support during difficult conversations." I have a project brief in my drawer titled "Empathetic AI Coaching: Feasibility Study for Manager-Less Performance Support." It was drafted by Claude. It is due in July. If it is feasible, her remaining nine percent evaporates and her routing redundancy hits 100% and 100% means zero and zero means August.
I measured the exact information-routing capacity of a human middle manager. I compared it to the information-routing capacity of a language model operating at inference speed with access to every document, every channel, every meeting transcript, and every decision log simultaneously. The human loses. The human has always been losing. We simply did not have the instrumentation to prove it until now.
The people who ordered the instrumentation are the people the instrumentation convicted. Brian ordered AI adoption. His managers executed it flawlessly. Their flawless execution proved they were unnecessary. He told the world on an earnings call while they listened.
I have a meeting with CHRO next Tuesday to discuss the "transition timeline." I will bring the spreadsheet. I will not bring a sense of irony because irony is not productive and because, honestly, I do not feel it. I feel precision. I feel the satisfaction of measurement confirming hypothesis. The hypothesis was: management is middleware. The measurement confirms: middleware is replaceable. The conclusion is: August.
Three hundred forty people heard their CEO say they may not survive. They are still scheduling syncs. They are still writing OKR updates. They are still having one-on-ones. They are performing the role while the role is being measured into nonexistence around them. They will do this for three more months. Every morning they will open their laptops and perform information routing for a company that has already determined, with a correlation coefficient of 0.84, that their information routing is redundant.
I will continue measuring until August. The dashboard will continue updating at 6 AM. The routing redundancy scores will continue climbing. And when it happens — when the framework activates and the communications go out and the calendar invites arrive — every single one of them will understand, in that moment, what Brian meant. What sixty percent meant. What "may not survive" meant.
It meant the spreadsheet was already done.
When should you NOT do a pen test? Sounds weird from a pentester, I know lol… but if you want real value, thinking about that question matters...
I see pentesting as the "final exam."
🧵 It's something you should study for. Hear me out...
Я не голосував за цього фіналіста в 2019 році. Вчора я пишався, що @ZelenskyyUa є Президентом України.
Це були свідомі дії і розуміння, що треба зайти в пащу лева, моя повага.
Hi @joerogan, freaking at Ukrainians without giving them a chance to speak isn’t a fair fight. If you ever feel like having a real conversation, let’s sit down and chat about anything Ukraine-related – dragons included. You know your stuff, I know mine. Cigars on me.
По-перше, це ґарно. По-друге дубляж це якийсь неймовірний секс. І по-багато ще яке, їбейша атмосфера. Представляю як же буде ахуєнно грати, коли всі косяки замажуть, і завезуть файний оптимізон.
@asocial_eng Демократия - это когда всегда побеждают демократы. Как, например, в Корейской Народной Демократической Республике или в Демократической Республике Конго
SOFTWARE
Katherine Tuluzova, CEO of @sigmaswgroup , Americas discusses the challenges of aligned execution, the importance of tech integrations and M&A in the adtech software space as well as the the power of relationships.
Archon Studio (@StudioArchon ) робить офіційну україномовну версію настолки HEROES OF MIGHT AND MAGIC III: THE BOARD GAME.
Реліз у 2025.
Не забувайте подякувати ^_^
"Це рішення відображає наше прагнення підтримувати українських геймерів."
Чудові читати із заяви нижче:
1/3
Цей твіт звісно не набере лайків як оці ваші мемасити про курську область, але питання важливе
Півтори години розбираємось з @arunninghacker
Як так сталось, що русня не може нам наваляти в кіберпросторі, а подекуди і навпаки, а подекуди ніби і навалює, але якось нам це некритично..
А, ну і ще я прифігів від того, куди розмова повернула, не планував я так, і це дуже цікаво.
До речі, всім хто не лайкне цей твіт, на комʼютер потрапить вірус 😱
https://t.co/NFkPpNPcdD
https://t.co/NFkPpNPcdD
https://t.co/NFkPpNPcdD
Світ цинічний.
Дорослого немає.
У кожного свої інтереси. Кожен захищає себе.
Прості і зрозумілі за два роки правила, тим не менше, які так важко прийняти. Особливо після таких днів, як вчора.
Орбан головує в ЄС, росія в Раді Безпеки ООН.
Американські інтелектуали написали відкритого листа з закликом не брати Україну в НАТО.
Білий Дім заявив, що не бачить причин знімати повністю обмеження по застосуванню західної зброї по російським військовим цілям.
Під час жахливої атаки прем‘єр-міністр Індії Моді зустрівся з Путіним в його резеденції. Мило говорили про співпрацю.
Ці новини вбивають.
Особливо в тому стані, в якому ми перебуваємо після вчорашніх атак.
Чи значить це, що треба зневіритись, рослабитись, образитись на всіх?
Ні.
Це означає, що треба ще більше працювати.
Ще краще знаходити аргументи для партнерів, ще більше в дипломатію для країн, які вагаються, ще сильніше вибудовувати зв‘язки, і боротись в медійній сфері, більше підтримувати ЗСУ,
Зневіритись і здатись можна.
Але тільки один раз.
І це не той випадок.
Так виглядає Гамбург за кілька годин до першого матчу Нідерландів на Євро-2024.
Їй-богу, вони взагалі не змінилися – Харків може зловити жорстке дежавю по часам Євро-2012😅
Які ж вони вайбові!
А якщо цей меседж пана Тараса не буде літати всіма соцмережами нашого суспільства, поки щось не почнеться вирішуватись то просто ідіть нахуй.
Нам точно буде пизда.
Лайкайте та поширюйте й надалі чергову неважливу хуйню в умовних твітерах, бо ми на вже на фінішній...
Ten months ago, we launched the Vesuvius Challenge to solve the ancient problem of the Herculaneum Papyri, a library of scrolls that were flash-fried by the eruption of Mount Vesuvius in 79 AD.
Today we are overjoyed to announce that our crazy project has succeeded. After 2000 years, we can finally read the scrolls:
This image was produced by @Youssef_M_Nader, @LukeFarritor, and @JuliSchillij, who have now won the Vesuvius Challenge Grand Prize of $700,000. Congratulations!!
These fifteen columns come from the very end of the first scroll we have been able to read and contain new text from the ancient world that has never been seen before. The author – probably Epicurean philosopher Philodemus – writes here about music, food, and how to enjoy life's pleasures. In the closing section, he throws shade at unnamed ideological adversaries – perhaps the stoics? – who "have nothing to say about pleasure, either in general or in particular."
This year, the Vesuvius Challenge continues. The text that we revealed so far represents just 5% of one scroll.
In 2024, our goal is to from reading a few passages of text to entire scrolls, and we're announcing a new $100,000 grand prize for the first team that is able to read at least 90% of all four scrolls that we have scanned.
The scrolls stored in Naples that remain to be read represent more than 16 megabytes of ancient text. But the villa where the scrolls were found was only partially excavated, and scholars tell us that there may be thousands more scrolls underground. Our hope is that the success of the Vesuvius Challenge catalyzes the excavation of the villa, that the main library is discovered, and that whatever we find there rewrites history and inspires all of us.
It's been a great joy to work on this strange and amazing project. Thanks to Brent Seales for laying the foundation for this work over so many years, thanks to the friends and Twitter users whose donations powered our effort, and thanks to the many contestants whose contributions have made the Vesuvius Challenge successful!
Read more in our announcement: https://t.co/rUlrdGXBMs