Abekek Notes

Continuing on that, if you’re a new grad at a Big Tech company (and I’m speaking from my experience at Amazon), it’s extremely hard to be a product owner. You won’t have the reputation or skills when you join. Product owners are usually Principals, Senior engineers, and a minority of mid-level engineers who can shape the product vision, lead development, and explore unknowns. Everyone else is responsible for smaller components where the decisions are already made, and in the era of AI that basically comes down to passing a spec to an LLM and letting it code for you. I got lucky because I joined Amazon at a pretty chaotic time. I was on 4 different teams and products in the span of 2 years, which sounds bad on paper, but it meant I got thrown into situations where things weren’t figured out yet. No clear ownership, no established hierarchy, just problems that needed solving. That’s where I got the chance to actually own a product, and it taught me more than any stable well-structured team would have. So if you’re early in your career, I’d recommend either joining a chaotic Big Tech team that doesn’t have its hierarchy figured out yet, or going to a smaller company like a startup where you can wear many hats and actually shape what gets built.

I keep seeing YouTube videos from laid-off software engineers, seniors with 10+ years, talking about how the field is dying, how coding doesn't bring joy anymore. If you loved the craft of coding, the problem solving, digging into edge cases, writing clean solutions, AI has taken a lot of that away. But I don't think the industry is dying. I think it's splitting into two subtypes. There's one type of engineer whose joy came from the process of writing code, where code itself was the output. For them, yeah, this feels like the end. Then there's another type who thinks more like a product owner, where code was never the point, the product was. They care about what gets built, not how, and for these people AI isn't a threat, it's what makes them faster. Be an owner, not just a coder.

Вчера сделал hard launch в Линкедине о том, где начал работать. К сожалению, больше ничего про работу говорить не могу

You can play around here: https://colab.research.google.com/drive/1WerUDMvfZmcwqeEwHTcccBP2NV6gtU9N?usp=sharing

I just spent my whole evening fixing BTS' "Into the Sun" from ARIRANG because the vocoder on V and Jungkook's vocals was driving me crazy. Basically what I did: downloaded the song, separated vocals from instrumentals with AI (Demucs), then used RVC (Retrieval-based Voice Conversion). It's an AI tool that converts one voice into another while keeping the melody and timing. so I found trained voice models for V, Jungkook, Jimin and Jin, ran the isolated vocals through each model which basically re-synthesizes the singing in their natural voice without the vocoder effect. Had to go frame by frame through a color coded lyrics video just to figure out who sings when so i could apply the right model to the right parts. Then blended the AI vocals with the originals and mixed everything back with the instrumentals. The result is not bad, but still has a bit of robotic voice 🤖

Отдых от работы не проходит зря, с другом опубликовали препринт идеи которая пришла в голову. AI-агенты всё чаще сами пишут себе инструменты и копят их в библиотеку. Но оценивают их только по тому, решили ли задачу. Это как судить разработчика только по "запустилось или нет", игнорируя repeated patterns, regressions и tech debt. Мы сделали бенчмарк который смотрит на качество самой библиотеки. Главный вывод: системы с почти одинаковым результатом (63–68%) отличаются по качеству кода на 18%. Стандартные метрики эту разницу не видят. Зачем это нужно: если агенты начнут массово генерить код в продакшене, нужно понимать не только "работает ли", но и насколько этот код поддерживаем и безопасен. https://arxiv.org/abs/2604.00392

Great talk from Prof. Kamilov https://youtu.be/fGuLlFGI2D8?si=BQ1pXX5LfL5E6O40

Сегодня был мой последний день в Амазоне. Bittersweet but excited about what’s next

I made a post on X yesterday and it kind of blew up. Lmk your thoughts and star if interested: https://github.com/abekek/arise

If I had a company, I would enforce no-AI days (maybe once a week), where all access to AI is restricted (unless you have some urgent launch or task). Costly, but the cost of your devs brain rotting might cost more.

Archaic is live on Product Hunt! Please upvote and comment: https://www.producthunt.com/posts/archaic-ai

Now Archaic is a fully redesigned desktop app, and integrates directly with your Claude/Kiro/Cursor workflow via MCP (bi-directionally): https://www.tryarchaic.com/ Download and check it out

Another cool product: https://www.pencil.dev/ Actually made Claude Code use their MCP to create me a design first

You guys can try it: https://testflight.apple.com/join/XjDjmVvv

Claude Code is doing magic: working iOS app + backend on AWS with infra and Strands SDK agent

I am doing apartment hunting, so did a quick MVP in 1 day to help me with that

Кстати, я сюда не писал, но около месяца назад — когда у Амазона были массовые сокращения — меня и всю мою команду тоже сократили. Но мне "повезло": менеджер, с которым я работал раньше, позвал к себе в команду, и я перешёл на новую роль внутри компании. Лишний раз убедился, что личные связи и репутация — это, наверное, самое важное в карьере. Позже напишу отдельный пост с выводами и тем, чему эта ситуация научила.

Check this out: https://www.tryarchaic.com/

Thanks, y’all! We got a lot of people that we interviewed, and were able to validate our idea. Glad to have such awesome and talented subscribers!