Software engineer, 20y+ years. Exploring AI in real work and everyday life. Coding agents, local AI on a 3070, and experiments worth sharing.
Stuttgart Area, Germany
- Tweets 6,788
- Following 1,139
- Followers 3,245
- Likes 16,458
Pinned Tweet
I'm not quitting my 9-5.
I'm building toward having that choice.
Exploring AI agents, local AI on my RTX 3070, and ideas that could earn on the side.
Sharing what works, what fails, and what I ship.
Anyone else figuring this out alongside a full-time job?
2
113
I think I pushed my hardware running local AI as far as I could (ryzen 5 5600x / 32gb ddr4 / rtx3070):
All qwen 35b a3b fine-tunes at 128k-150k context:
• 1800 t/s peak prefill
• 40 tok/s peak generation
Do you think I should get more? (MTP was not making things faster)
1
73
What did make you change your mind about AI coding?
DHH traces his shift in confidence through several generations of models, including a period of disappointment.
Have your own results changed your view?
For me it was Opus 4.5 in November 2026
2
181
AI still amazes me:
Debugged an app that was slower despite having more hardware.
Code, Query Plans, Indices, Latency on the cluster.
DB was in the wrong availability zone (eu west vs ger west).
Took me 10 minutes with an agent. Other team looked at me like a wizard.
1
130
People including myself will quote this a lot over the next months
It's pencils down, people. Writing code by hand is no longer an economically viable skill for most programmers at most companies. But the future of making software has never been brighter. Don't you dare black pill this beautiful moment! piped.video/vDjW_dRyKXY?si=6Fsf…
1
2
203
Still not over the fact we won't have qwen 4 35b a3b and I won't have the compute to run qwen 4 27b 😔
1
2
205
Looking forward to gpt-6-luna the most, the model that carries me through my corporate job all day long 😂
We might be getting a DOUBLE DROP today.
Opus 5.5 is dropping. Now GPT 6 Sol, GPT 6 Luna, and GPT 6 Astra Minor just showed up in Microsoft's Azure config overnight.
OpenAI demoted GPT 5.6 Sol to "workhorse" in the same commit. You do not do that unless the replacement is ready.
We are finally about to get intelligent models we can actually use on our subscriptions.
The biggest day in AI this year might be today.
2
1
238
Time for hot takes:
How would a world look like where Frontier Intelligence wouldn't have a token/month limit?
1
132
My 3070 is crying for mercy, but qwen 2.1 image works on bf16 😆 Unbelievable what @QwenDevs
cooked here, feels like nanobana2 running on my crappy 8gb of VRAM
192
I think at some point I would just be happy for my RTX 3070 to die so I can get something with more VRAM ...
6
6
499
Everytime I am over getting a GPU ... something new drops
Meet Qwen-Image-2.1, the most balanced and cost-effective image generation model in the Qwen-Image series! Now open weights! 🎨
A unified model for both generation and editing, delivering top-tier quality in a lightweight package.
Highlights: 👀
- Compact & exceptionally fast: A lightweight 7B architecture that outperforms most closed-source models, with drastically accelerated inference for multi-image inputs.
- Native transparency: Natively generates and edits RGBA layers, enabling seamless compositing and text editing within transparent images.
- Versatile, high-fidelity editing: Supports up to 10 reference images and precise local control while preserving strict fidelity for portraits and products.
- Broad coverage & stunning aesthetics: Excels at panoramas, infographics, and virtual try-ons, delivering realistic textures and elegant typography.
Start to create your next masterpiece with Qwen-Image-2.1! 🖼️
- Blog: qwen.ai/blog?id=qwen-image-2…
- GitHub: github.com/QwenLM/Qwen-Image…
- Model Scope: modelscope.cn/models/Qwen/Qw…
- Hugging Face: huggingface.co/Qwen/Qwen-Ima…
2
218
Used Codex to:
• not getting scammed while buying a Asus Ascend
• update my Linkedin profile with computer use which used my full 5h window ... lol
How is your day going so far?
247
I am also a bit puzzled by all these founders "you can now classify things, my whole dataset with 5M entries"
... could have build your own classifier by spending $5 to generate a dataset out of the data
Jev model is very polarizing in my social vicinity: people who got exposed to AI after ChatGPT are positively giddy with excitement, like they have literally just discovered fire. Pre-GPT ML people are baffled how that could've made news at all :-)
2
4
605
realized people are not aware what you can run on 8gb vram + 32gb of ram
All the well known 35b a3b models (qwen, ornith, etc) at 4bit
Here are a few commands to get you started:
gist.github.com/eiselems/e65…
~/Work/llama-cpp/llama.cpp/build/bin/llama-server \ -m ~Qwen3.6-35B-A3B-UD-Q4_K_M.gguf \ -fa on -ctk q4_0 -ctv q4_0 \ -b 8192 -ub 4096 \ --no-mmap --jinja -t 6 \ 1.5 \ -c 131072 --fit on -np 1 \ --host 127.0.0.1 --port 8080
1
439