*https://www.youtube.com/watch?v=JKk77rzOL34
**https://300.ya.ru/v_pC71wTmG
таймкоды
00:00:00 Рекорд автономного кодирования
- Агенты с ИИ установили рекорд по продолжительности автономного кодирования — две недели.
- Создан полностью функциональный компилятор языка C, более 100 строк кода на Rust.
- Компилятор соответствует требованиям специального набора тестов на 99%.
00:00:58 Быстрое развитие ИИ
- За год автономное программирование с ИИ увеличилось с 30 минут до двух недель.
- Исследователи удивлены скоростью изменений.
- Опус 4.6 выпущен 5 февраля, версия cutting edge кажется устаревшей.
00:01:54 Улучшения в контекстном окне
- Опус 4.6 имеет контекстное окно в миллион токенов, что в пять раз больше, чем у предыдущей версии.
- Возможность хранения 50 строк кода в одной контекстной сессии.
- Значительное улучшение поиска документов в кодировщике.
00:02:43 Взаимодействие агентов
- Команды агентов работают автономно, обмениваясь сообщениями.
- Реальное автоматическое взаимодействие между агентами в корпоративной системе.
- Инструменты января уже устарели.
00:03:43 Управление разработчиками
- ИИ может управлять 50 разработчиками одновременно.
- Важен тест MRCVR2 score для измерения способности модели извлекать информацию из длинного контекстного окна.
- Опус 4.6 показывает 76% вероятность найти «иголку в стоге сена» при миллионе токенов.
00:05:39 Целостное понимание системы
- Модель может удерживать всю систему целиком, включая импорт, зависимости и взаимодействия между модулями.
- Старший инженер имеет ментальную модель всей системы, что отличает его от подрядчика.
- Опус 4.6 анализирует контекст аналогично человеческому разуму.
00:06:32 Будущее ИИ
- Рабочая память значительно улучшилась за пару месяцев.
- Проект компилятора языка C требовал 16 параллельных агентов, но это временно.
- Rakuten внедрил облачный код в свою инженерную организацию, используя Опус 4.6.
- Опус 4.6 закрыл 13 выпусков в системе отслеживания проблем Rakuten.
00:07:22 Автоматизация управления проектами
- Искусственный интеллект распределил 12 выпусков среди 50 человек за один день.
- Модель анализирует не только код, но и организационную схему, понимая зависимости между командами и инженерами.
- Автоматизирована координация работы, на которую инженеры-менеджеры тратят половину времени.
00:08:21 Экономическая эффективность
- Работа старшего инженера в Rakuten может стоить до четверти миллиона долларов в год.
- Opus 4.6 демонстрирует возможность автоматизации оперативной координации, которая обычно занимает 15–20 часов в неделю.
- Многочасовое автономное кодирование становится обычным явлением.
00:09:21 Параллельная работа агентов
- Разрабатывается агент, который разбивает сложные задачи на 24 параллельных сеанса работы с облачным кодом.
- Нетехнические сотрудники Rakuten могут вносить вклад в разработку через интерфейс терминала Cloud Code.
- Граница между техническим и нетехническим трудом стирается.
00:10:17 Командные рои агентов
- Opus 4.6 использует командные рои агентов для координации работы.
- Агенты работают независимо и могут напрямую связываться друг с другом.
- Создан компилятор языка C с помощью 16 параллельных агентов.
00:11:40 Иерархические структуры в ИИ
- Автономный рой агентов самостоятельно организовался в иерархические структуры.
- Иерархия — это возникающее свойство координации действий множества интеллектуальных агентов.
- Управление — это то, что делает интеллект для координации действий в масштабе.
00:13:37 Обнаружение уязвимостей
- Opus 4.6 обнаружил более 500 ранее неизвестных уязвимостей высокой степени серьёзности.
- Уязвимости были проанализированы исследователями в области безопасности и автоматизированными инструментами.
- Код, прошедший аудит, содержит множество уязвимостей, требующих немедленного исправления.
00:14:09 Новый подход к анализу кода
- Модель самостоятельно анализирует историю Git проекта, читая многолетние журналы коммитов.
- Она определяет области, где изменения, связанные с безопасностью, были внесены поспешно или не полностью.
- Использует понимание эволюции кода для поиска уязвимостей, недоступных статическому анализу.
00:15:06 Мыслительная модель модели
- Модель создаёт мысленную модель кода, а не сканирует известные шаблоны.
- Исследует слабые места с изобретательностью исследователя и терпением машины.
- Релиз Opus 4.6.0 представлен как обновление для нового поколения.
00:16:03 Скептицизм и изменения в моделях
- Скептицизм по поводу новых выпусков моделей ИИ прослеживается исторически.
- Опытные пользователи могут быть разочарованы изменениями в рабочих процессах.
- Выпуски моделей часто включают компромиссы и изменения в систему управления агентами.
00:17:45 Примеры использования ИИ
- Репортёры CNBC с помощью ИИ создали замену инструменту управления проектами Monday.com за час.
- ИИ может создавать рабочие версии приложений за день без необходимости написания кода.
- Персональное программное обеспечение становится всё более доступным.
00:19:41 Новый ритм работы
- ИИ позволяет выполнять несколько задач за несколько минут.
- Пользователи описывают результаты, а не процесс, что меняет подход к работе.
- Переход от операционных инструментов к управлению агентами.
00:20:41 Ясность намерений
- Главное сейчас — ясность намерений, а не технические навыки.
- Разработчикам нужно точно формулировать требования для агентов ИИ.
- Маркетинговым командам важно знать, какие показатели важны, и объяснять их значение.
00:21:39 Экономические преимущества ИИ
- Компании, использующие ИИ, показывают высокий доход на сотрудника.
- McKinsey прогнозирует увеличение количества агентов ИИ в компании к 2026 году.
- Организационная структура компаний меняется, уступая место управлению агентами ИИ.
00:22:40 Стартапы и искусственный интеллект
- Джейкоб Бэнк проводит маркетинговую операцию стоимостью в миллион долларов с нулевым количеством сотрудников и примерно сорока агентами ИИ.
- Компания Micro One проводит три собеседования с использованием ИИ каждый день.
- Три разработчика в Лондоне создали полноценную платформу для бизнес-банкинга за шесть месяцев, на что до появления ИИ потребовалось бы двадцать инженеров и восемнадцать месяцев.
- Новая модель стартапов: два-три человека плюс целый штат специализированных агентов, организованных по результатам.
00:23:36 Организационная схема и роль людей
- Организационная схема становится картой команд агентов-людей, каждый из которых владеет полным рабочим процессом.
- Вопрос для руководителей: сколько агентов на человека — правильное соотношение?
- Важны отличное суждение, понимание потребностей клиента и знание предметной области.
00:24:29 Перспективы стартапов
- Дарио Амаде оценивает шансы на то, что к концу 2026 года компания, основанная одним человеком, заработает миллиард долларов, в пределах 70–80%.
- Сэм Альтман делает аналогичные ставки среди руководителей технических компаний.
- Связь между количеством сотрудников и объёмом производства нарушена, организации, которые первыми определят новое соотношение, опередят конкурентов.
00:25:31 Будущее ИИ и инфраструктура
- Агенты разрабатывают полноценные приложения в течение месяца или более за раз.
- Инвестиции в гипермасштабируемую инфраструктуру стоимостью в 650 миллиардов долларов выглядят консервативными.
- Центры обработки данных строятся для азиатских компаний, работающих в таких масштабах, что людям было трудно их смоделировать или осознать.
00:26:27 Адаптация к изменениям
- Попробуйте запустить мультиагентную сессию на реальной работе, чтобы изменить своё представление о возможностях агентов.
- Измените ментальные модели для работы с ИИ.
- Менеджеры должны оценить, сколько времени команда тратит на оперативную координацию, и понять, какую часть работы ИИ может взять на себя.
00:28:22 Напряжённость и адаптация
- Агенты ИИ могут создавать компиляторы производственного уровня за две недели, управлять инженерными подразделениями автономно и обнаруживать сотни уязвимостей в системе безопасности.
- Всё идёт быстрее, и будущее неопределённо.
- Лидерам нужно время, чтобы помочь своим людям пережить переходный период.
00:30:20 Развитие мышления
- Каждый месяц важен для развития мышления в контексте ИИ.
- Будущее работников умственного труда зависит от их способности идти в ногу со временем и работать с агентами ИИ.
In this video
Chapters
Transcript
16 Agents Coded a C Compiler in Two Weeks
0:00
Claude Opus 4.6 just dropped and it changed the AI agent game again because
0:05
16 Claude Opus 4.6 agents just coded and set the record for [snorts] the length
0:11
of time that an AI agent has coded autonomously. They coded for two weeks straight. No human writing the code and
0:18
they delivered a fully functional C compiler. For for reference, that is over a 100,000 lines of code in Rust. It
0:25
can build the Linux kernel on three different architectures. It passes 99% of a special quote torture test suite
0:32
developed for compilers. It compiles Postgress. It compiles a bunch of other things. And it cost only $20,000 to
0:38
build, which sounds like a lot for you and me, but it’s not a lot if you’re thinking about how much human equivalent
0:43
work it would cost to write a new compiler. I keep saying we’re moving fast, and it’s even hard for me to keep
0:50
up. A year ago, autonomous AI coding could top out at barely 30 minutes before the model lost the thread. Barely
0:56
30 minutes. And now we’re at 2 weeks. Just last summer, Rock 10 got 7 hours out of Claude and everybody thought it
1:03
was incredible. 30 minutes to 2 weeks in 12 months. That is not a trend line.
1:08
That is a phase change. The entire world is shifting. Even one of the anthropic researchers involved in the project
1:15
admitted what we’re all thinking. I did not expect this to be anywhere near possible so early in 2026. Opus 4.6
1:24
shipped on February 5th. It has been just over a week. And the version of cutting edge that existed in January,
30 Minutes to Two Weeks in 12 Months
1:30
just a few weeks ago, that already feels like a lifetime ago. Here’s how fast things are changing. Just in Anthropic’s
1:38
own road map, Opus 4.5, shipped in November of 2025, just a couple of months ago. It was Anthropic’s most
1:44
capable model at the time. It was good on reasoning, good at code, reliable against long documents. It was roughly
1:50
the state-of-the-art. Just a few months later, Opus 4.6 6 shipped with a 5x
1:55
expansion in the context window versus Opus 4.5. That means it went from 200,000 tokens to a million. Opus 4.6
2:03
shipped with the ability to hold roughly 50,000 lines of code in a single context
2:09
session in in its head, so to speak, up from 10,000 previously with Opus 4.5.
2:15
That is a 4x improvement in coder document retrieval over just a couple of months. The benchmarks measures are off
2:22
the charts. And you guys know I don’t pay a lot of attention to benchmarks, but when you see something like nearly
2:28
doubled reasoning capacity on the ARC AGI2 measure, you got to pay attention. It shows you how fast things are moving,
2:34
even if you don’t entirely buy the benchmark itself. And Opus 4.6 adds a
2:39
new capability that did not exist at all in January. Agent teams. Multiple
2:45
instances of cloud code autonomously working together as one with a lead agent coordinating the work. specialist
2:52
handling subsystems and direct peer-to-peer messaging between agents. That’s not a metaphor for collaboration.
Opus 4.6: 5x Context Window Expansion
2:58
That is automatic actual collaboration between autonomous software agents in an
3:03
enterprise system. All of this in just a couple of months. The pace of change in AI is a phrase that people keep
3:09
repeating and they don’t really internalize what it means. This is what it means. The tools that you mastered in
3:15
January are a different generation from the tools that shipped this week. It’s not a minor update, people. It is an
3:22
entirely different generation. Your January mental model of what AI can and cannot do is already wrong. I was
3:28
texting with a friend just this past week, and he was telling me about the Rockin results in 7 hours. And I had to
3:35
tell him, I know you think you’re up to date, but the record is now 2 weeks. And by the way, Rockuten using Opus 4.6 was
3:43
able to have the AI manage 50 developers. That is how fast we’re moving. that AI can boss 50 engineers
3:49
around. Now, the 5x context window is the number anthropic put in the press release. It’s the wrong number to focus
3:55
on. The right number is a benchmark originally developed by OpenAI called the MRCV2
4:02
score. That sounds like a mouthful and it’s used to measure something that matters enormously that nobody was
4:07
testing properly. Can a model retrieve and use the information inside a long context window? In other words, can you
4:14
find a needle in the haststack? It’s not about whether you can quote unquote put a million tokens into the context
4:19
window. Every major model can accept big context windows in January 2026. The
4:25
question is whether the model can find, retrieve, and use what you put in there. That is what matters. Sonnet 4.5, which
4:31
was a great model from Claude just a few months ago, does have a million token window, but the ability to find that
4:38
needle in the haststack was very low. About one chance in five or 18.5%. Gemini 3 Pro a little bit better at
4:45
finding that needle in the haststack across its context window about one chance in four 26.3%.
4:51
These were the best available in January. They could hold your codebase. They couldn’t reliably read it. The
4:57
context window was like a filing cabinet with no index. Documents went in, but retrieving them was kind of a random
The Real Number: Needle-in-Haystack Retrieval
5:04
guess past the first quarter of the content. Guess what? Guess what? Opus 4.6 six at a million tokens has a 76%
5:12
chance of finding that needle in the hay stack. At 256,000 tokens or a quarter of the context window, that rises to 93%.
5:20
That is the number that matters. That is why 4.6 feels like such a giant leap. It’s not because of the benchmark score.
5:27
It’s because there’s a massive difference between a model that can hold 50,000 lines of code and a model that
5:33
can hold them 50,000 lines of code and know what’s on every line all at the same time. This is the difference
5:40
between a model that sees one file at a time and a model that holds the entire system in its head simultaneously. Every
5:47
import, every dependency, every interaction between modules, all visible at once. A senior engineer working on a
5:54
large codebase carries a mental model of the whole system and they know that changing the O module can break the
5:59
session handler. They know the rate limiter shares state with the load balancer. It’s not because they looked it up. It’s because they’ve lived in the
6:05
code long enough that the architecture becomes a matter of intuition, not a matter of documentation. That holistic
6:11
awareness is often what separates a senior engineer from a contractor reading the codebase for the first time.
6:16
Opus 4.6 can do this for 50,000 lines of code simultaneously. Not by summarizing,
6:23
not by searching and not with years of experience. It just holds the entire context and reasons across it the way a
6:29
human mind does with a system it knows very very deeply. And because working memory has improved this dramatically in
6:35
the span of just a couple of months, it’s actually not hard to see where the trajectory is going to go from here. The
6:40
C compiler project 100,000 lines in Rust did require 16 parallel agents precisely
6:46
because even a million token context window can’t hold that whole project at once in its head. But at the current
6:52
rate of improvement, it won’t require 16 agents for long. Let me tell you more about the Rockuten story with the 50
6:58
developers. Now, Rocket is a Japanese ecom and fintech conglomerate, and they deployed clawed code across their
Holistic Code Awareness Like a Senior Engineer
7:04
engineering org, not as a pilot, but in production, handling real work and touching real code that ships to real
7:11
users. Use Kaji, Rakuten’s general manager for AI, reported what happened when they put Opus 4.6 on their issue
7:18
tracker. Clawed Opus 4.6 closed 13 issues itself. It assigned 12 issues to
7:24
the right team members across a team of 50 in a single day. It effectively managed a 50 person org across six
7:30
separate code repositories and also knew when to escalate to a human. It wasn’t
7:36
that the AI helped the engineer close the tickets. I want to be clear about that. It closed issues autonomously. It
7:41
did the work of an individual contributor engineer. It also routed work correctly across a 50 person org.
7:48
The model understood not just the code but the org chart. Which team owns which
7:53
repo? which engineer has context on which subsystem, what closes versus what needs to escalate. That’s not just code
8:00
intelligence, that is management intelligence. And a system that can route engineering work correctly is a
8:06
system that understands organizational dependencies the way a human lead understands them. Which means the
8:12
coordination function that engineering managers spend half their time on just became automatable in a couple of
8:18
months. Think about the cost structure that implies. A senior engineering manager at a company like Rakuten might
8:25
cost a quarter million dollars a year fully loaded, maybe more. A meaningful part of their job, ticket triage, work
8:32
routing, dependency tracking, cross team coordination. That is exactly what Opus
8:37
4.6 demonstrated it could handle. Not the judgment calls about what to build next, not the career development
Rakuten: AI Managing 50 Developers
8:42
conversation, and it wasn’t done over weeks and weeks and weeks, but the fact that it can do operational coordination
8:49
that typically takes 15 to 20 hours a week and demonstrated it could do it for a full day, it shows you where things
8:56
are going. And the broader numbers tell the same story. It is common now to see hours and hours and hours of sustained
9:02
autonomous coding for individuals who are playing with this. not in the controlled enterprise environment even
9:07
people can kick off multi-hour long coding sessions and just walk away and do other things and come back and see
9:14
fully working software. That is no longer an unusual thing in February 2026. And Rockwood isn’t stopping here.
9:21
They’re building an ambient agent that breaks down complicated tasks into 24 parallel clawed code sessions. Each
9:28
single one handling a different slice of their massive mono repo. A month of human engineering is generating a
9:35
simultaneously running 24 agent stream that helps them to build and catch issues and that’s in production. Now the
9:42
detail that gets buried under these big numbers might be more interesting than all of the numbers themselves because
9:48
non-technical employees at Rockoten are able to use that system to contribute to
9:53
development through the cloud code terminal interface. That is right. The terminal is not just for engineers
10:00
anymore. People who have never written code are able to ship features because of the work Rockin has done to integrate
10:06
cloud code. So the boundary between technical and non-technical, it keeps breaking that down. The distinction that
10:11
has organized knowledge worker hiring and compensation for 30 years is dissolving in a matter of months. It’s
10:18
not dissolving at the speed of your ability to deploy a multi-month project and is not dissolving at the speed it
10:25
takes to retrain humans. That is why this is shocking. This is all happening faster than we can adjust to it. One of
10:31
the features that is most hard for us to wrap our minds around is the agent teams feature that Opus 4.6 shipped. Tropic
10:38
calls them team swarms internally, which is a little scary and I can see why the marketing team changed that. But the
10:43
name is accurate. It’s not a marketing term. It’s an architecture. Multiple instances of clawed code are architected
10:50
to run simultaneously. Every single one in its own context window. And they coordinate through a shared task system
10:56
that has three simple states, right? Pending, in progress, and incompleted. Pending, in progress, and completed. One
11:02
instance of Cloud Code is going to act as your lead developer. It will decompose the project into work items and assign them to specialists, track
11:09
dependencies, and unlock bottlenecks. This is just like what Opus 4.6 did for
11:14
those 50 developers. The specialist agents work independently, and when they need something, they don’t just go
11:20
through the lead, by the way. They can message each other directly. peer-to-peer coordination. It’s not hub and spoke. There’s a front-end agent,
11:27
there’s a back-end agent, there’s a testing agent. Effectively, they are recreating the entire software
11:32
engineering org inside claude code team swarms. And this is how that C compiler got built. It’s not one model doing
11:39
everything sequentially, right? It’s 16 agents that worked in parallel. Some building the parser, some building the
11:45
code generator, some building the optimizer. And they all coordinated through the same kinds of structures
11:50
that existing human engineering teams use, except they work 24 hours a day.
11:56
They don’t have standups, and they resolve coordinations through direct messaging rather than waiting for the
12:01
next sprint planning session. One of the running questions in AI has been whether agents will reinvent management. I think
12:07
this argues strongly that they did. Curser’s autonomous agent swarm independently organized itself into
12:13
hierarchical structures and strong DM published a production framework called software factory that’s built around
12:18
exactly the same hierarchical pattern. And now anthropic has shipped a feature with 13 distinct operations from
12:24
spawning managing to coordinating agents. This is not really coincidence. It’s essentially convergent evolution in
12:30
AI. Hierarchy isn’t a human organizational choice imposed on systems
12:36
to ma imposed on systems to maintain control. It’s an emergent property of
12:41
coordinating multiple intelligent agents on complicated tasks. Humans invented management because management is what
12:47
intelligence does when it needs to coordinate at scale. AI agents effectively discovered the same thing
12:52
because the constraints are structural. They’re not cultural. You need someone to track dependencies, right? You need
12:58
specialists. You need communication channels. You need a shared understanding of what has been done and what hasn’t yet been done. We did not
13:04
impose management on AI. AI effectively discovered management and we helped to
Agent Teams: Hierarchy as Emergent Property
13:10
build the structure and Opus 4.6 is the first model that ships with the infrastructure to run all of this as
13:16
just another feature. On the same day Opus 4.6 launched, Enthropic published a result that got much less attention than
13:23
that C compiler story, but it might matter more in the long run. They gave Opus 4.6 six basic tools, Python,
13:29
debuggers, fuzzers, and they pointed it at an open-source codebase. There were no specific vulnerability hunting
13:36
instructions. There were no curated targets. This wasn’t a fake test. They just said, «Here’s some tools. Here’s
13:42
some code. Can you find the problems?» It found over 500 previously unknown
13:48
high severity, what’s called zeroday vulnerabilities, which means fix it right now. 500 in code that had been
13:55
reviewed by human security researchers scanned by existing automated tools
14:00
deployed in production systems used by millions of us. Code that the security
14:06
community had considered audited with when traditional fuzzing by the way
14:11
fuzzing is the fancy technical word for finding bugs and making sure you check all the code. It’s like fuzzing your hand through the carpet and finding a
14:18
pin or something. And when manual analysis failed, using a tool called ghost script, which is what you use to
14:25
check these things, the model independently decided to analyze it a different way, going directly to the
14:30
project’s git history. That’s right. It worked around obstacles and it read through years of commit logs to
14:36
understand the codebase’s evolution. Nobody told it to do this. It just decided to do it. And it identified
14:42
areas where security relevant changes had been made hastily or incompletely all on its own. It invented a detection
14:49
methodology that no one had told it to use. It reasoned about the code’s history, not just about its current
14:56
state. And it used that understanding of time to find vulnerabilities that static
15:01
analysis could not reach. Humans didn’t do this. This is why it found the bugs it found. This is what happens when
15:08
reasoning meets working memory. The model doesn’t scan for known patterns the way existing tools do. It builds a
15:13
mental model and I think that’s the only metaphor that works at this point of how the code works, how data flows, where
15:20
trust boundaries exist, where assumptions get made and where they might break down. And then it probes the
15:26
weak spots with the creativity of a researcher and the patience of a machine that never gets tired of reading commit
15:32
logs. And I guarantee you, human engineers get tired of that. The security implications alone would
15:38
justify calling Opus 4.6 a generational release. And yet again, I remind you, it’s only been a couple of months.
15:44
There’ll be another one in a couple of months. But this was not the headline feature of Opus 4.6. As exciting as it
15:49
is, it wasn’t even the second headline feature. 500 zero days was the side
15:54
demonstration. That is the density of capability improvement that has been packed into a single model update
500 Zero-Day Vulnerabilities Found Autonomously
16:01
shipped on a Wednesday, February. Look, there are skeptics for every release and there were skeptics for 4.6 as well. And
16:08
the skepticism tracks historically. AI benchmark improvements have underdelivered before and repeatedly for
16:14
years. And that is why I don’t depend a lot on benchmarks. Sure enough, within hours of launch, thread started to
16:20
appear on the cloud subreddit. Is Opus 4.6 labbotomized? Is it nerfed? The
16:26
pattern seems to repeat with every major model release. Power users who have fine-tuned their workflows for the
16:32
previous version discover the new version handle certain tasks differently. The Reddit consensus, for what it’s worth, has decided that 4.6
16:38
six is better at coding and worse at writing. I haven’t found that personally, but dismissing it would also
16:44
probably be dishonest. Model releases involve trade-offs. Model releases often involve changes to the agent harness,
16:51
which is all of the system prompting that the model makers put in that goes around the deployment that they don’t
16:57
talk about and they don’t release. We don’t know how it changed. We feel the change when we work with the system. So,
17:02
I’m sure it’s possible that if you were a Reddit user who was used to a special prompt pattern that worked on Opus 4.5,
17:10
you might indeed be frustrated that that pattern did not work on a much more capable model overall. So, I get the
17:16
skepticism. I also get that people are tired of hearing the world changing every couple of months. It is exhausting
17:22
to keep up with. But that doesn’t mean it’s not real. And that is part of why I’m telling so many specific stories.
17:28
It’s important not to just look at the headlines. It’s important not to look at some number changing on some stupid
17:33
benchmark. It’s important to hear the stories of how AI is actually changing in production now. So what does this
17:40
feel like if you are not an engineer? What does this feel like if you don’t write code? Cuz the C compiler, let’s be
17:46
honest, it’s a developer story. The benchmarks are developer metrics. But the change underneath, what makes 4.6
17:54
special isn’t about developers per se. It’s about what happens when AI can sustain complicated work for hours and
18:00
days instead of minutes. Two CNBC reporters, Dear Drabosa and Jasmine Woo, they’re not engineers, right? They’re
18:06
reporters. They sat down with Claude Co-work and they asked it to build them a Monday.com replacement. That’s the
18:12
project management tool, right? A project management dashboard that had calendar views. It had email integration, task boards, team
18:19
coordination features. This is the product that monday.com has spent years and hundreds of millions of dollars
18:25
building. It s currently supports a $5 billion market cap for monday.com. It
18:31
took these reporters under an hour. Total compute cost somewhere between $5 and $15. I hasten to add that is not the
18:39
same thing as rebuilding monday.com. This was personal software. It’s not deployed. It’s not for sale. It was just
18:46
for them. So yes, it is a big deal. Yes, it is a generational change in our
18:51
ability for nontechnical people to make software. No, I am not saying that dear
18:57
Drebosa and Jasmine Woo can refound monday.com for $10. The real story is
19:02
that AI can build the tools you use, the software you pay per seats for, the dashboards your company spent 6 months
19:09
specking out with a vendor, an AI agent can build a working version of that in an afternoon, and you don’t need to
19:15
write a line of code to make it happen. Yes, it might just be for you. It’s a whole new category, people. It’s called
The Skeptics and Reddit Reactions
19:21
personal software. It didn’t exist just a few months ago. It is now increasingly easy to make that happen. Our daily
19:28
experience with AI is changing in ways that are really difficult to benchmark, but that doesn’t mean that they’re not
19:34
structural. A marketing ops team using Claude Co-work can do content audits in
19:39
just a few minutes instead of hours and hours. A finance analyst running due diligence doesn’t take a day to do it
19:46
because the model can read the document set, identify the risks, and produce lawyer ready redlines in just a few
19:52
minutes. Our rhythm of work is different now. We can dispatch five different
19:57
tasks in a few minutes on Claude Co-work. We can dispatch a PowerPoint deck, a financial model, a research
20:03
synthesis, two data analyses, right? walk away, you can grab a cup of coffee, and you can come back and the
20:09
deliverables are just done. They’re not drafts anymore. It’s just finished work.
20:15
It’s mostly formatted, right? The pattern that’s emerging for non-technical users is what anthropic
20:20
Scott White calls quote vibe working. You describe the outcomes, not the process. You don’t you don’t tell the AI
20:26
how to build the spreadsheet. You tell it what the spreadsheet needs to show. It figures out the formulas. It figures
20:33
out the formatting. It figures out the data connections. The shift is coming for all of us and it’s going from
20:39
operating tools to directing agents. And the skill that matters now is not technical proficiency. It’s clarity of
20:46
our intent. Knowing what you want, being able to articulate the real requirement,
20:52
not just your surface request. That is becoming the bottleneck. Ironically, it’s the same bottleneck the developers
20:58
are hitting, but from a different direction. The C compiler agents didn’t need anyone to write code for them. They
21:04
needed someone to specify what a C compiler means precisely enough that 16 agents could coordinate on building one.
21:11
The marketing team doesn’t need someone to operate their analytics platform anymore. They need someone who knows
21:16
which metrics matter and can explain why. The leverage across the board has shifted from execution to judgment
21:23
across every function. Whether you write code or not, if you lead an organization, the number that should
Non-Engineers Building Software in an Hour
21:28
restructure your planning isn’t measured in weeks or days. It’s actually measured in revenue per employee. Purser, the AI
21:35
coding tool hit $und00 million in annual recurring revenue with about 20 people.
21:40
That’s $5 million per employee. Midjourney generated 200 million people.
21:45
Midjourney generated $200 million with about 40 people. Lovable, the AI app builder, they reached $200 million in 8
21:52
months with 15 people. For traditional SAS companies, $300,000 in revenue per
21:58
employee is considered excellent and $600,000 is considered elite. That would
22:03
be notion. AI native companies are running at five to seven times that number. Not because they found better
22:09
people necessarily, but because their people orchestrate agents instead of doing the execution themselves. McKenzie
22:15
published a framework not for others, but for themselves last month. They’re targeting parody, matching the number of
22:22
AI agents at McKenzie to human workers across the firm by the end of 2026. This
22:28
is the company that sells organizational design to every Fortune 500 on Earth. And they’re saying the org chart is
22:35
about to flip. The pattern is visible at startups, too. Jacob Bank.
22:40
The pattern is visible at startups, too. Jacob Bank runs a million-doll marketing operation with zero employees and
22:47
roughly 40 AI agents. Micro One conducts 3,000 AI powered interviews every single
22:52
day, handled at a tiny fraction of the headcount that enterprise recruiting firms need floors of people to do. Three
22:59
developers in London built a complete business banking platform in 6 months. A project that would have required 20
23:06
engineers in 18 months before AI. Amazon’s famous two pizza team formula, the idea that no team should be larger
23:13
than what two pizzas can feed, is evolving into something even smaller. The emerging model at startups is now
23:19
two to three humans plus a fleet of specialized agents all organized not by function but by outcome. The humans
23:26
regardless of what their job title says and it increasingly doesn’t matter set direction, evaluate quality and make
Vibe Working: Describing Outcomes, Not Process
23:32
judgment calls. The agents execute, coordinate, and scale. The org chart stops being a hierarchy of people and it
23:39
becomes a map of human agent teams each owning a complete workflow end to end.
23:45
For leaders, this changes the fundamental equation we’ve been working with for a long time. It’s not about how
23:50
many people do we need to hire now. It’s about how many agents per person is the right ratio and what does each person
23:57
need to be really excellent at to make that ratio work. The answer to the second question is really the same thing
24:03
that’s distinguished really excellent people for a long time in software. It’s great judgment. It’s what we call taste,
24:09
which is vague, but typically means deeply understanding what the customer wants and being able to build it. It’s
24:15
about domain expertise. It’s the about the ability to know whether the output is actually really, really good. And
24:21
those skills now have 100 orex leverage because they are multiplied by the number of agents that person can direct
24:28
and drive against. out. Daario Amade, anthropic CEO, has set the odds on a billion dollar solo founded company
24:35
emerging by the end of 2026 at between 70 and 80%. Think about it. He thinks
24:41
there’s a 75% chance that there will be a billion dollar solo founded company by
24:46
the end of this year. Sam Alman apparently has a betting pool among tech CEOs on the exact same question. Now,
24:53
whether or not you believe that version, the direction is undeniable. The relationship between headcount and
24:59
output is broken. And the organizations that figure out the new ratio first are going to outrun everybody who is still
25:06
assuming they need dozens of developers to do one major software project. If you follow the trajectory that Opus 4.6 set,
25:14
by 2026, June, July, August, somewhere in there, I would expect agents working
25:19
autonomously for weeks to become routine rather than remarkable. By the end of
25:24
the year, we are likely to see agents building full applications over potentially a month or more at a time.
25:31
Not toy applications, real production systems with real architecture decisions complete with security reviews, with
25:38
test suites, with documentation, all handled by agent teams. The trajectory from hours to two weeks took just 3
25:45
months. The trajectory from two weeks to months, that’s coming soon. And the inference demand that this generates
25:51
agents consuming tokens continuously around the clock across thousands of parallel sessions. Companies are not
Revenue Per Employee at AI-Native Companies
25:57
ready for this. This is what makes the $650 billion in hypers scale infrastructure look conservative rather
26:04
than insane. Those data centers are not being built for chatbots people. They’re being built for agent swarms running at
26:10
a scale that people have had difficulty modeling or wrapping their heads around. Opus 4.6 gives us a sense of that
26:17
future. So, what can you do about it? If you’re sitting here thinking, «Oh my gosh, this is too much. It is coming too
26:23
fast.» You’re not alone. You’re not alone. You can do this. If you write
26:28
code, try running a multi- aent session on real work, not a toy problem, a piece of your codebase with real technical
26:34
depth. Watch how the agents coordinate. That experience is going to change your mental model of what agents can do in a
26:41
way that matters much more than anything else. Because increasingly the way we work is the bottleneck for AI. If we
26:48
want to go faster and build more, if we want to feel like we have the ability to
26:54
do production work at the speed that AI demands because increasingly that’s what they’re going to expect from humans. I
27:00
got to say, if we want to be ready for the future, best way we can do it is to change our mental models. If you don’t
27:07
write code, open up Claude Co-work. handed a task you’ve been procrastinating on that’s that’s felt
27:12
really hard, right? A competitive analysis task, maybe a financial model task, a content audit across last
27:18
quarter’s output. Just describe the outcome you want, not the steps to get there. See what comes back. The gap
27:24
between what you expect and what you get is the gap between your current mental model and where the tools are today. And
27:31
for managers, look honestly at the 20 hours a week your team spends on operational coordination, ticket
27:37
routing, dependency tracking. ask how many of those hours really require excellent human judgment and which are
27:43
just pattern matching because I got to say your AI can probably take over a lot
27:48
of the coordination work already and if you run an organization if you’re on the senior leadership team you got to
27:54
understand the question for your org has changed it’s not about should we adopt AI or even which teams adopt it first
28:01
it’s really what is our agentto human ratio and what does each human need to be excellent at to make that ratio work
28:07
and how do We support our humans to get there. The people working in knowledge work desperately need their leaders to
28:14
understand that humans need a ton of support to get through this change management and become a new kind of
28:21
worker that partners with AI. That is not an easy thing and most orgs are underinvesting in their people. I tell
28:27
people in AI that if you are on the cutting edge of AI, it feels like you’re time traveling always because you look
28:34
at what’s happening around you and then you go and you talk to people who haven’t heard about it and they look at
28:39
you like you’re crazy. They say, «No, you can’t do that. You can’t run 16 agents at a time and run a Rust
28:45
compiler. What do you mean?» And AI can manage 50 people. And when you tell them that’s just a Wednesday in February and
28:52
more is coming soon, then they really roll their eyes. But welcome to February. This is where we are. AI
28:58
agents can build production grade compilers in just two weeks. They manage engineering orgs autonomously. They can
29:04
discover hundreds of security vulnerabilities that human researchers missed. They can build your competitor’s product in an hour for the cost of your
29:11
lunch. They can coordinate in teams, resolve conflicts, and deliver at a level that did not exist 8 weeks ago.
29:17
None of this was possible in January. And we don’t know where this stops. We just know it’s going faster. That’s the
29:24
tension underneath all of the benchmark scores, all the deployment numbers. The fact is the agents are here, they work,
The Billion-Dollar Solo Founder Prediction
29:30
and it’s just getting faster from here. And we’re not sure what happens next. The question I have for all of us is how
29:39
do we do a better job supporting each other in adjusting to what feels like a race out of MadMax some days? Welcome to
29:46
February. It’s moving fast. If you’re a people leader, you need to take time to
29:51
think about how to support your people to make it through this transition. If you are an individual contributor or a
29:56
manager, I am putting as many tools as I can up on the Substack to help you get
30:01
through this. But the best thing you can do, it’s not about the Substack. I don’t care. It’s about you touching the AI and
30:09
getting hands-on and actually building or trying to build with an AI agent
30:15
system that launched not in January, not in December, but in February. And you
30:20
need to take that mindset forward every single month. In March, you should be touching an AI system that was built in
The Trajectory From Here
30:27
March. Every month now matters. Make sure that you don’t miss it because our future as knowledge workers increasingly
30:34
depends on our ability to keep the pace and work with AI agents.