Соавтор статьи — Сергей Левенец, CTO в команде ТестОпс.

Что останется делать людям, когда код будут писать машины? Практика показывает — на удивление много, без работы никто не останется. В двух статьях мы опишем реальный опыт создания внутреннего агента для исправления багов в команде ТестОпс. Расскажем, сколько такой агент экономит, как его отлаживать, и чем автоматическая правка отличается от диалога с Claude Code.

В команде ТестОпс разработка поставила себе задачу: ноль багов в бэклоге. Эта амбициозная задача запустила большую работу по исследованиям и разработке, в результате которой родился агент, способный исправлять и проверять баги. Команда стихийно назвала его «Агент Смит».

Написать первую версию агента было делом двух-трёх дней. Гораздо сложнее оказалось внести в него экспертизу — то есть отладить его шаги и настроить помощников. Для этого потребовалось и набить шишек в реальной работе, и изучить много научной литературы. Возможно, самым сложным и важным было научиться оценивать издержки от внедрения агента. Об этом мы и поговорим — но вначале разберёмся, как устроен сам «Смит».

Mount Etna has continued to erupt in Sicily, sending molten lava, rocks and ash down its slopes as hikers watch nearby. The Italian volcano is Europe's most active

Youngest White House press secretary in history says she wants more time to ‘be the best mom’. Plus, are we losing the battle against AI slop in the music charts?

• Don’t already get First Thing in your inbox? Sign up here

Good morning. A familiar face in Donald Trump’s administration is departing, as the US president announced that Karoline Leavitt was stepping down to spend more time with her family. The president said Leavitt, who became the youngest person to serve as White House press secretary on his return to office in January 2025, “will now be one of my top outside advisers”.

Trump’s most prominent spokesperson is departing while his approval rating is at a historic low, as voters express discontent over rising prices and a war in Iran that has no end in sight.

Why did Leavitt say she was stepping away? Leavitt wrote: “The truth is since returning to the White House after the birth of my daughter, I have felt in my heart that I cannot be the best mom my two young children deserve while devoting the constant time, energy and attention required of the White House press secretary – and that is why I have ultimately made the bittersweet decision to depart the White House and embark on a new chapter in my life.”

How will Leavitt’s time in the role be remembered? At 27 when she was appointed, Leavitt was the youngest White House press secretary in history. And when she announced she was pregnant in December last year, she became the first ever expectant press secretary. She has faced criticism for saying “Gen Z and those younger than me have been raised with just silver spoons in their mouths”, and for attacking reporters trying to hold the White House to account.

What are conditions said to be like onboard? In April, there was a flurry of public criticism about dire conditions onboard the carrier. Photos emerged of poor-quality and rationed food, and there were reports of water shortages, moldy showers and broken laundry machines. Pete Hegseth, the defense secretary, dismissed the accounts as “more fake news”.

How long has the ship been at sea? The warship left San Diego on 21 November and was originally intended for the Pacific but was redirected to the Middle East once the US-Israel war with Iran began. Since then, the 5,000-plus crew has had only two days on land. The deployment was scheduled to end in May but has been extended amid Trump’s ongoing war.

Предположим, вас укусил комар прямо в задницу.

Что там, вы не видите, но зудит, поэтому вы чешетесь. То есть прям берёте и начинаете водить ногтями по месту укуса. До тех пор, пока это отвратительное состояние зуда в моменте не пройдёт.

Прикол в том, что почесаться надо достаточно сильно, чтобы боль от царапин перекрыла сигнал зуда в спинном мозге. Тогда вы чувствуете облегчение на несколько секунд. Но механическое воздействие на это место усиливает выброс гистамина и цитокинов, и раздражение становится сильнее. А раздражение — главная причина зуда.

Есть ещё невероятный прикол с серотонином в этой же цепочке, но про него чуть позже.

Боль вообще-то — технология отрицательной обратной связи. Не делай, чтобы тебе не было больно. Не трогай горячее; не выгибай вот так руку; береги пораненную ногу; не бей организмом по кокосу, бей лучше камнем и так далее.

Зуд от укуса комара даёт положительную обратную связь. Чем больше зудит, тем больше зудит.

Часто бывает, что 90% итогового воспалительного процесса — это не комар и его токсины, это мы своими ногтями.

Соответственно, если вы хотите, чтобы что-то на коже заживало, нужно успокоить пользователя, отключить чувствительность в первую очередь. Потом убрать раздражение на уровне сигнальных молекул (загасить цитокиновый микрошторм). Потом уже можно заниматься починкой структуры и полирнуть сверху защитным барьером, потому что сухой роговой слой сам по себе раздражает рецепторы.

И вот у нашей лаборатории появилась задача это всё собрать. Опыт за 30 лет накопился, в регенерации тканей мы уже хорошо разбираемся, но прошло не без сюрпризов.

12 августа SpaceXAI выпустила Grok 4.6. Модель получила 500 тысяч токенов контекста, четыре режима рассуждения и явный уклон в длинную агентную работу. В первых тестах она держится рядом с сильнейшими конкурентами, а в CursorBench выходит вперед при средней стоимости 2,81 доллара за задание.

Цифра выглядит приятно. Почти так же приятно, как счет из ресторана, пока никто не вспомнил про вино. У Grok 4.6 таким вином стал длинный контекст: после 200 тысяч токенов тариф удваивается. Поэтому разберемся, где модель действительно прибавила, сколько стоит длинный прогон и кому есть смысл пробовать ее прямо сейчас.

The 2026 midterm cycle is projected to be the most expensive election cycle in US history, with candidates across the board raising money at a record pace. And voters are taking notice: Big-donor groups tied to issues like Israel policy, AI, and crypto have become major flashpoints in primary races in both parties.

Campaign spending alone doesn’t determine elections. The recent Democratic Senate primary in Michigan was the most expensive Democratic congressional primary in terms of outside spending — but ended with Abdul El-Sayed overcoming a major financial disadvantage to win the nomination. In California, billionaire Tom Steyer spent over $200 million of his own money on his bid for governor, but lost. 

But the ever-rising tide of cash helped fuel voter concerns about corruption and the campaign finance landscape is only getting more confusing: In June, the Supreme Court struck down limits on how much political parties can spend in coordination with their own candidates, further blurring the line between what candidates raise directly and what parties can spend on their behalf.

How can voters follow the money? How much difference does outside spending actually make in the outcomes of these races? And how much can candidates rely on grassroots donors to push back? 

Danielle M. Thomsen, a professor of political science at the University of California Irvine and author of the 2025 book, The Money Signal: How Fundraising Matters in American Politics, has spent years studying these questions. The conversation has been lightly edited and condensed for clarity.

Let’s start with a race like the recent Michigan Senate Democratic primary. Abdul El-Sayed was massively outspent by outside interest groups like AIPAC and still won. Does that change how you think about money’s role in politics, or is it generally consistent with what we know?

The bulk of my research is looking at US House primaries, and I think you could probably draw similar conclusions for Senate primaries, except that in most cases, Senate primaries have way more information. So in the case of Michigan, I think people knew largely what the choices were. And after you get to that stage, largely through money and advertising and voter outreach, and campaigning, after that point, money might matter less because voters become more familiar with choices. But this is a really steep hurdle in most congressional primaries.

I think that people forget how low information these contests usually are. Many primary voters really don’t know much about candidates. I think in the Michigan Senate example, that was less true. Money isn’t irrelevant, but when you have other pieces of information, that can override money. And furthermore, in that case, both of the candidates were spending millions of dollars in the primary. I mean, this is just so unique in a lot of ways.

Is there a minimum level of money congressional candidates need to be viable? Do we know?

The average for non-incumbent successful candidates in open seat primaries is about three to $400,000 in the first quarter. And if you look at incumbents who lost or struggled, well, their challengers for the most part, are not, you know, broke jokes. They’re raising $200,000 to $300,000 in their first quarter, which resembles what quote unquote “open seat candidates” are raising too in many districts. Maybe not your super wealthy districts like in New York or in California, but in many districts, $300,000 in your first quarter is a really good showing. So I would say money is money, and early money in particular is a prerequisite.

And that’s not going to mean that if you raise a lot of money, you definitely will win. That’s never been the case and that’s still not the case. But it does matter for you, it matters for who is seen to be seriously in the mix, who is perceived to be relevant. And in this cycle, just like in most cycles, virtually all contenders who are either successful or are seen as successful show support through early fundraising.

Your view is that fundraising works largely as a signal of viability and strength — to the public, party leaders, media, everyone. Is that the consensus view in political science, or are there competing frameworks?

In many ways the argument of my book is not very groundbreaking if you talk to practitioners, but it hasn’t been empirically documented in the ways that I did. Some other people in political science call attention to endorsements. The other theory is that money matters not as a signal of viability but for material goods — so to buy advertisements, office infrastructure, hiring staff, and consultants.

Within a candidate’s total haul, does it matter where the money comes from, like a broad base of small donors, a candidate self-funding, or a super PAC backed by a few wealthy people?

I’ve recently begun to look at this. With a graduate student of mine, Ryan Mundy, we collected data and wrote a paper on why wealthy candidates are likely to win. Because it’s not just the case that rich candidates are more likely to run, which most people know and prior research has shown, but among those who run, rich candidates are also more likely to win.

And what we show is that rich candidates are more likely to raise early money. In particular, they’re more likely to raise large-dollar donations, which then have a greater return in subsequent quarters.

What do we know about candidates who fully self-fund?

So I’ve learned that fully self-funded candidates, meaning those who don’t supplement with financial support from others, typically lose. Self-funding on its own is not as valuable as getting money from others, and in particular large-dollar donations from others.

However, many, many candidates supplement with their own money. Something like 40 percent of non-incumbent general election winners self-fund at least $10,000. I think people don’t really understand the prevalence of self-funding as a supplement, and how that keeps some people disadvantaged, particularly those who can’t self-fund at all.

For some self-funders, I think some of them just don’t want to raise money. It can be pretty unpleasant! And while self-funding with no supplemental money from others is overwhelmingly a bad strategy, there are a couple handful of people who have done this and won out of our sample that looked at 2014 to 2024. There were 407 non-incumbent general election winners in that period. And 17 of them did bankroll at least 90 percent of their first-quarter dollars and won. So not very many, but there are a handful of exceptions who do it and win, and they are all rich. So funding is complicated.

And on the “friends and family” side of early money, who’s actually writing those first checks?

Almost all early money comes from individual contributors. Most of that is itemized, so contributors over $200. Most of our elections are driven by the upper class. On average non-incumbent general election winners are getting like 50 percent of their early money from max-out donors, meaning people who pay the maximum of $6,600. Who can afford to pay that? Not normal Americans. It’s basically politically engaged, politically connected upper-middle class people who know people who run for office.

And the other thing is most of them are not repeat donors! Most of them don’t even give to more than one candidate. Because they’re probably just giving to the one rich person they know who is running for office that year. That’s why in political circles the first quarter is called the “friends and family quarter” — that’s when political consultants tell candidates to open up their Rolodex and call everybody they know from when they started kindergarten to who their coworkers are now.

Is there a point, empirically, where more spending stops helping or even backfires? I’m thinking of a case like Tom Steyer in California.

So the assumption has always been that money has diminishing returns. That goes back to Gary Jacobson’s work from forever ago. But nobody in political science puts a number on it. What they do is they say things like, Oh, the relationship is positive until at some point it goes down.

When outside groups pour money into a race, like AIPAC now, or the charter school lobby and the NRA in earlier cycles, can that spending become politically useful to the candidate it’s targeting, from backlash or otherwise?

Yeah, so I don’t think there’s actually good research on this. I don’t think there’s enough information. But I think in general most people don’t know where money comes from. So the backlash is minimal. I’ve asked candidates who raised a lot of money from self-funding, “Did your self-funding ever come up?” And they’d say, “No, it never came up. I thought it would either help me, to show that I wasn’t beholden to special interests, or it would hurt me because I didn’t have the support that having fundraising money from others would demonstrate. But it didn’t come up.” Another candidate who raised a lot of money out of state told me that while it was raised a few times by people on Twitter, no one in person ever asked him about it.

Many people don’t know their own members of Congress. So you can imagine why there wouldn’t be a backlash, because people don’t know enough for there to be one — except in these cases where it does become an issue, like the recent Michigan Senate Democratic primary. It’s certainly part of the narrative for candidates like Elizabeth Warren, her stance against big money is a big part of her persona, same with Bernie Sanders and AOC. But those are real celebrity candidates.

In the 2010s, small-dollar fundraising was talked about as a democratizing counterweight to big money, but I remember reading that research found it could also make politics more polarized. Where did that political science literature land, and has the rise of small donors changed anything as Super PACs and mega-donors have grown more powerful?

So I would say the research is exactly how you describe it. There are so many studies that have shown that small-dollar donors tend to be more liberal Democrats and more conservative Republicans. So they are coming from the ideological edges of the spectrum. And there’s a lot of evidence of that. So that’s the finding, that’s the literature.

If you look at how candidates raise money by and large today, it’s not coming from small donations or unitemized donations. Most candidates are raising the bulk of their money from itemized donors. And most money is still from large donors.

So your read is that, despite all the attention on small-dollar fundraising, the underlying picture hasn’t shifted much either direction?

There’s a tendency to look at the cases that don’t conform to the rule rather than the rule. Like Eric Cantor [who lost in 2014] and AOC [who won in 2018]. In some cases, the losers raised way more than the partisanship of the district in those cases would predict, and then you have candidates who raised no money and unpredictably won. So there’s always this tendency to prop up the exceptions to the rule. In political science, though, we like to look at the rule.

The dangers of AI become clearer every day. Why are we still acting as if we have no choice about our future?

Rather than producing jobs, the US economy actually lost 23,000 jobs in July, according to Bureau of Labor Statistics data released on Friday. In addition, May and June’s job numbers were revised downward, showing a combined 103,000 fewer jobs than previously reported.

As if this weren’t bad enough, wage growth has also slowed. Average hourly earnings rose by just 0.1% from June.

Привет, Хабр! Меня зовут Эрнест Гилязов, я тимлид группы разработки интерфейсов Рунити. В этой статье расскажу, как мы переносим на новый стек проект Рег.ру, кодовой базе которого исполняется 20 лет, и почему в итоге Nuxt-приложение само раздает шапку и футер в легаси-монолит через обычный JSON-эндпоинт.

Global food prices have hit their highest level in over three years. We are told this is due to rising oil and energy costs, but how exactly do they affect prices? The most reported reason is that transport and fertiliser are to blame, but the reality is that the oil and gas industry is far more integrated in our food production than many of us realise. To get a better understanding, Josh Toussaint-Strauss takes a closer look at the production processes of three categories of food: meat, processed staples and fruit and vegetables. What he discovers is a level of involvement that is shocking

Всем привет! 

Меня зовут Дмитрий Шипилов, с вами команда поиска Uzum Market.

Мы уже рассказывали вам о нашем поисковом движке: о том, как вообще работает ранжирование и что у него под капотом и как мы получаем кандидатов с помощью векторного поиска. 

Но хороший поиск — это не только правильное ранжирование и мощная retrieval-часть. Хороший поиск — это набор различных инструментов и компонентов, собранных в сложную систему.  Поэтому в этой статье вы узнаете о важных вещах, без которых наш поиск не может обойтись, в том числе: как мы помогли опечаточнику понимать контекст нашего маркетплейса, почему безобидные синонимы могут привести к неожиданным эффектам, в каких случаях важно сужать количество товаров и как построить устойчивую LLM-систему для оценки качества ранжирования. 

Почти любой NVMe-диск сегодня выпускается в двух версиях: с радиатором и без. Разница в цене между ними выходит не такая уж и большая, поэтому многие не глядя берут более дорогую. Да и чего бы не взять? Доплата копеечная, зато на выходе получаешь диск, который точно не перегреется. Звучит логично, вот только чаще всего деньги улетают в трубу. Не потому, что есть сценарии, когда SSD не греется. Греется. Просто настоящая причина ставить радиатор лежит совершенно в другой плоскости, о которой почему-то почти не говорят.

Critic of Consumer Financial Protection Bureau leadership was put under investigation and suspended from job

The Trump administration has been accused of “bullying and intimidation” after a labor union leader and critic was put under investigation and suspended from his job at the Consumer Financial Protection Bureau (CFPB).

Stephen Wheeler, a data scientist at the agency since 2022 and chair of the CFPB union organizing committee, was placed on administrative leave effective immediately earlier this month. The union said it was not provided with any specific violations or given any details of the investigation.

Two years after southern Brazil’s worst climate disaster ravaged the city, and with poor areas still recovering, many fear official neglect has left them exposed once again

In the backstreets of Porto Alegre, a state capital in southern Brazil, you can still see signs of the devastating flood that swept through the city two years ago: scattered debris, waterlogged homes and destroyed businesses.

In May 2024, flooding killed 183 people in the state of Rio Grande do Sul and displaced more than 600,000. The neighbourhood of Sarandi was one of the hardest-hit parts of Porto Alegre, with water levels reaching up to 5 metres above normal, completely submerging many of its streets.

Weaker growth, rising energy costs and persistent inflation mean Britain could face tougher times in the autumn

• UK economic growth slows down

• Business live – latest updates

In the spring the International Monetary Fund warned Britain faced the heaviest economic blow from the Iran war among the world’s most advanced nations. Almost six months into the conflict, on the surface, the UK appears to be proving the forecasters wrong.

The latest official figures show the UK maintained its pole position as the fastest growing economy in the G7 in the first half of 2026. Despite the gloomy international backdrop and yet more domestic political uncertainty, consumers have largely continued spending and business investment has boomed.

Споры между Elementor и Bricks обычно сводятся к попытке определить победителя: что быстрее, удобнее, чище, функциональнее.

После нескольких лет работы с WordPress мне такой вопрос кажется не совсем правильным.

Я много работаю с Elementor PRO и не собираюсь от него отказываться. При этом для новых проектов всё чаще смотрю в сторону Bricks.

Причина не в том, что один билдер «хороший», а второй «плохой». Они по-разному вписываются в процесс разработки.

Для себя я сейчас разделяю их примерно так:

Распределенная система может сломаться так, что по отдельности все ее части будут выглядеть исправными.

Представим обычную оплату заказа: сервис отправил запрос платежному провайдеру, тот списал деньги и вернул успешный ответ. На обратном пути соединение оборвалось. Для платежной системы операция завершена, а сервис заказа получил тайм-аут и не знает, произошло списание или нет.

Самое очевидное решение — повторить запрос. И получить второе списание.

В монолите подобные ситуации встречаются реже: операция обычно проходит внутри одного процесса или одной транзакции, поэтому место сбоя проще определить. В распределенной системе между началом и концом одной бизнес-операции могут быть несколько сервисов, брокер сообщений, разные базы данных и внешний API (интерфейс программирования приложений). У каждого компонента при этом свое состояние и свое представление о том, что уже произошло.

Так, проверки «отправили запрос — получили ожидаемый ответ» здесь недостаточно. Интереснее проверить, что будет, если ответ задержится или потеряется, запрос придет повторно, события поменяются местами, а один из сервисов восстановится после нескольких минут простоя.

В таких ситуациях проявляются ошибки, которые сложно увидеть на happy path (позитивном сценарии). Ниже разберем, как их воспроизводить и что проверять, чтобы отказ одного компонента не превращался в некорректное состояние всей системы.

Почему распределенную систему нельзя тестировать как обычное приложение

Возьмем простой пример — оплату заказа в интернет-магазине.

For Democrats, impeachment remains an option, but they are reluctant to make it a centerpiece of their agenda this time.

More than 500 people have lost jobs at FedEx, Staples and LAZ Parking; two facilities in Palm Springs and Victorville will close.

У меня хостинг-провайдер, в штате шесть человек. Когда я говорю, что мы внедрили СЭД, знакомые ржут: вам-то зачем, вас на одну переговорку хватит.

Пробежимся по цифрам. Шесть сотрудников и под сотню точек доступа, больше половины которых ведет не ко мне, а к клиентам. Учет всего хозяйства я вел в гугл-таблице. Заблудились мы в ней, когда нас было трое. Не шестеро, а трое.