Sebastian Heitmann

tl;dr accomplished developer and consultant. Writing about tech, entrepreneurship and mental health. Last year at this time, I officially started my solopreneurship. This was an important moment for me. I closed the chapter about my burnout and the following depression. It wasn't easy, but I landed a client, and I was able to sustain myself. I had many different goals, but as usual, it took longer than expected to achieve them. Now, my professional site is live, and I have started writing actual content. I am still developing my product ideas. This is nothing singular or astonishing in itself. I know that. But overall, I am very glad to have started posting here, and I am still around, so expect more to follow. Feel free to follow and tell me what you are up to! cya ✌️

@ZuiLoong @kritikakodes Oh, I get that, but your comment was phrased unreasonable hostile

@e2e_developer @kritikakodes Read my response again in retrospect to the original post

Genuine question: Why we are using python for building LLM, why not Java, if java is faster than python?

@ZuiLoong @kritikakodes Nor is the JVM. You can write wrappers for native bindings for the JVM. Python is easy to use and for academic people that have no background in software engineering a simpler tool. So it comes down to language.

@kritikakodes Tell me you know nothing about how computers actually work without telling me you know nothing about how computers actually work... P.S. language isn't the bottleneck there.

@kritikakodes Because scientists are not software engineers. They need a language that is easy to use and has good support for what it needs to do. By chance, that was Python.

@dev_maims At the moment, any job is hard to get.

Unpopular opinion: Software engineering is the only career where getting the job is harder than doing the job.

@ImLunaHey AnotherBrowser.

i need a name for the browser im working on, anyone got any ideas?

@Saanvi_dhillon It runs WSL.

other than gaming, give me ONE reason to buy a Windows laptop in 2026 rather than a Macbook??

@codevsdev Reading documentation. You spend more time understanding and researching what you are doing than actually building. With AI, this even becomes more important.

What's one tech skill you think is underrated?

@CreeCoder Weapons.

AI should never be used for replies. What are some other things AI should never be used for?

Hey people, I like to challenge my current dev workflow and see if I can create 5 viable landing pages in 5 days. If you have any recommendations for what I should create the landing pages for, let me know. You have two days, before I start. Have a nice weekend ✌️

Mediocre code does work. But working is just a sufficient baseline property. The problems occur when a product has a long lifecycle, and development loses momentum due to the code base's quality. AI will improve, but it doesn't really understand the semantics behind the code. It only predicts what should come next.

@e2e_developer Don't you think that AI will get better an better with code over time thanks to A players helping to train it? At the end of the day if mediocre code works and does it's job, does it really matter that it's mediocre? Given it's secure of coarse

tl;dr learn how to build software I want to be frank. AI writes mediocre code. It's flawed. But, from what I know by experience, it's the same mediocrity you find out there. I am fascinated by AI because it's not worse of what I've seen in usual software projects. The truth is we have more C players than A players around. We need to make sure that we implement practices, so that AI doesn't get away with writing bad code. I am doing a QA sprint for one of my projects at the moment. The good thing is, that AI often sees that something is off. But it often doesn't see the issue. That's where a senior operator is needed. So, you don't need to learn DSA to become good with agents. But you need to learn how to continuously deliver software and how to keep it from breaking

@FrancescoCiull4 @Abhishekcur AI is a tool, and it's still up to the artisans to decide how to use the tool. I am noticing how my experience and knowledge shape the quality of what I build with AI.

This si one of the topics I studied a lot during my CS degree. I mean, you cna learn mostof the things by yourself but most of the people jjust don't do that. And I strongly believe that even in the AI era, fubdamentals are fundamental, as they give you a clear vison of what the hell you are doing

if you're serious about systems engineering, don't skip operating systems and computer organization & architecture. they teach you what's really happening inside the machine how cpus execute instructions, how memory behaves, and where performance actually comes from. everything else starts making more sense after that. and the more i learn, the more i realize this is the true core. it's not just a syllabus thing. it's the foundation of understanding how computers actually work. that's why i keep coming back to these two subjects again and again. every revisit reveals something new, and every deep system suddenly makes a little more sense.

@Abhishekcur Very cool stuff. Wish I had time to dive into low-level coding.

been diving deep into cuda kernels lately, and it's honestly a lot of fun. one thing that's become very clear: understanding gpu architecture and learning tools like nsight compute, nsight systems, and ptx/sass using godbolt inspection is where real performance engineering starts, literally (it's heart). it literally connects you to the hardware and low-level system internals. every optimization feels like a small puzzle, and that's what makes it soooo enjoyable. performance isn't learned by reading alone, it's learned through profiling, measuring, debugging, and endless iteration, keep on working on this iteration loop. so, expect something on this too from me, hehe :) i love cpu's and gpu'ss, they are super tasty 😋😋

@petarivanovv9 @systemdesignone Agreed. It often comes down to the harness you use and how well you define goals.

Note: Before you design agents, tune prompts, or pick a model, understanding how to structure what goes into the context window is the most impactful skill in practical AI engineering. Most quality problems in production AI systems trace back to poor context design, not the wrong model or missing features.

If you want to get good at AI engineering (in 2026), learn these concepts: 1 LLM Evals Explained ↳ newsletter.systemdesign.one/p/llm-evals 2 Design Knowledge Q & A System ↳ newsletter.systemdesign.one/p/ai-based-kno… 3 How OpenClaw Works ↳ newsletter.systemdesign.one/p/openclaw-arc… 4 AI Agent Workflow ↳ newsletter.systemdesign.one/p/how-do-ai-ag… 5 How MCP Works ↳ newsletter.systemdesign.one/p/how-mcp-works 6 Design AI Chat Assistant ↳ newsletter.systemdesign.one/p/ai-chat-assi… 7 How RAG Works ↳ newsletter.systemdesign.one/p/how-rag-works 8 Agentic Patterns Explained ↳ newsletter.systemdesign.one/p/agentic-desi… 9 AI Coding Workflow 101 ↳ newsletter.systemdesign.one/p/ai-coding-wo… 10 Machine Learning System Design 101 ↳ newsletter.systemdesign.one/p/machine-lear… 11 Multi-Agent Architecture Explained ↳ newsletter.systemdesign.one/p/multi-agent-… 12 How AI Agents Work ↳ newsletter.systemdesign.one/p/how-do-ai-ag… 13 How Vector Databases Work ↳ newsletter.systemdesign.one/p/what-is-a-ve… 14 AI Agents: Memory, State & Consistency ↳ newsletter.systemdesign.one/p/ai-agent-mem… 15 AI Agents Design ↳ newsletter.systemdesign.one/p/ai-agents-ex… 16 Context Engineering 101 ↳ newsletter.systemdesign.one/p/what-is-cont… 17 What is Reinforcement Learning ↳ newsletter.systemdesign.one/p/what-is-rein… 18 LLM Concepts - A Deep Dive ↳ newsletter.systemdesign.one/p/llm-concepts What else should make this list? === 👋 PS - Want my System Design Playbook (for free)? Join my newsletter with 201K+ software engineers now: → newsletter.systemdesign.one/join === 💾 Save & RT to help others get good at AI engineering. 👤 Follow @systemdesignone + turn on notifications.

@_devJNS I tell people I am self-employed... 🙃

a good way to put it actually.

@catalinmpit I say, this is for the better, but it really shows how many coders don't actually understand either.

Code implementation is cheap now. The focus is now on architecture and system design, even more than before.

@thdxr People consider being autistic cool? Okay... You can be autistic and cool, by all means. But those aren't even correlated.

tech industry has put way too low value on being cool people even act like what they imagine autistic people are like everything could be so much easier and smoother if this culture was corrected

@__roycohen outcome > speed for me. The reason I stick to Claude is mostly reliability at the moment. I am considering trying out new models, including Gemini, after finishing my current projects.

Gemini is still on another level when it comes to perf/speed compared to anything else Factor of 10x in terms of speed..

@systemdesignone I got myself a Pixel 10 Pro. No particular reason.

TECH TWITTER ONLY Which Android phone would you recommend buying in 2026?

@Raynerdtech In my opinion, degrees are overrated. It is an achievement, but it doesn't say anything about your capabilities beyond.

How many people are actually working in the exact field they studied in school? It feels like everyone I know graduated in one thing and ended up building a career in something completely different.