Şenol Çolak

@denizirgin That is true, alignment to the structure problem is a reality. Specs and prompts make things easy. You always need to know what you are asking for.

I have ADRs, specs, a knowledge base, and structured skills, supported by a very lean AGENTS.md file that guides and onboards LLMs into projects. It explains the use cases, when to use each skill, which ADRs to read first, the overall guidelines, and so on. Even with all of that in place, I still need to put in additional effort to keep them on track.

Over the last 2-3 weeks, I've been working intensively with AI agents on long-running projects that involved multiple iterations and progressed in phases. My takeaway is this: AI coding tools have the knowledge of a senior developer, but the know-how of a junior developer. They know a lot: frameworks, patterns, edge cases, best practices, and so on. But they don't have production intuition, codebase hygiene, maintainability reflexes, or the feeling of "will this decision come back to bite us in 6 months?" That’s why strict code review, continuous standardization, guardrails, and clear direction are essential. Otherwise, it's very hard to maintain a high-quality, sustainable, and human-maintainable codebase. For months, I couldn't fully understand the whole "vibe coding" and "AI slop" discussion. Now I get it. If you leave AI on its own in a long-running project, it can really wreck your codebase. I guess that's why almost all demos are one-shot greenfield project demos😀

Yes, sure there will be source code. But machine code only.

@barisakin Güzel görünüyor

Oyuncak hali burada. Cok kisitli featurelarda tutmak istiyorum, belki birkac olmazsa olmaz feature eklenebilir ama sadece perf fix olacak. github.com/parisxmas/pyx

(convential coding) Data Structures -> (agentic coding) context management

@gorkemcetin Bence açık kaynak lisanslama modelleri şuanki problemler için yetersiz. Örn. Cal .com yerine @OlivierLamber12 cal.rs yazdı. Geliştirmeyi sadece boş zamanlarında yazmış, AGPL-3 veya Apache-2 bu yazılımları geliştirenleri korumak için yeterli değil.

Open source çok acayip bir yerden geçiyor. Ekonomik modelde bir kırılma var. Bir kaç örnek: - curl: AI tarafından üretilen hatalı güvenlik raporları geliştiricisini o kadar rahatsız etti ki bug bounty programını kapattı. - tldraw: Dışarıdan gelen PR'lar tamamen otomatik kapatılmaya başlandı. Yani katkı üretmek artık mümkün ama kabul edilme ihtimali neredeyse sıfır. - Cal .com: Açık kaynak kodunun AI ile daha hızlı taranıp exploit edilmesi riski nedeniyle kodunu kapatma kararı aldı. Geliştiricisi güzel bir video yayınladı, onu seyretmenizi öneririm. Sonuç olarak open source bitiyor mu, hayır. Ama daha kapalı, daha kontrollü ve kimin ne yaptığının bilindiği bir yapıya evriliyor.

As usual.. they are smart fellas💡 So far this is the best for us too

The Shift to AI Capital: Why We Need a New Breed of Collaboration Tools linkedin.com/pulse/shift-ai…

Companies are shifting from Human Capital -> AI Capital The real advantage will be how accurate and detailed companies can store knowledge and use that knowledge autonomous execution and feed back their AI again. @ycombinator AI native companies 👇 youtu.be/EN7frwQIbKc?si…

Small is always good (almost) 👍👇👇

Workaround is the workaround for the workaround that started with a workaround. - Engineering cult proverb

💡👇👇👇

@gokhanamal Harika bir yazı olmuş 👍. Özellikle 2+ ekipler için paylaşımlı yapı kurmak çok önemli

Şirket içinde shared bir yapı kurup, mühendisler arasında context paylaşımı yapmak agentic yazılım geliştirme konusunda fark yaratıyor. Bunu nasil daha kolay yapabileceginizi anlatan bir makale hazirladim 👇

@fkadev Data centric software is the key. Whatever you build, it should be generating a valuable data which can be stored. The abstraction can change

Coding is not the main skill right now in the AI era. The main skill is the ability of "verifying the future-proof software architecture"

💡👇

@ibuildthecloud The value was never the code. Reaching o(n) is not human. Having the expertise in a domain is the value. I personally asked that question "what was the craziest thing in my professional career that I said, woohoo it is technically very difficult to do" now we are focusing on them

I'll be kind of honest, it sorta sucks that everyone can barf code now. I'll get used to it, but it was nice to have a skill people couldn't easily reproduce in the worst way possible. I'm sure I have something valuable to offer, but code isn't it anymore.

🤔

@ibuildthecloud "build the cloud" . If software becomes a commodity, what won't change is the right question. My idea, The foundational services. DB, below and above that area which will provide the foundation services to the new AI era.

I'd like to do something useful.

@ibuildthecloud use spec-kit or bmad, otherwise %100 you will get lost.

Production quality vibe coding, that's the challenge.

This is a real game changer..👇