Franz und Franz

Programmer working on making myself obsolete · Hannover · AI enthusiast · Franz und Franz web agency · stargazer 🔭

[Franz und Franz @FranzUndFranz]

Re the @huggingface benchmark: 2026 update of an old rule: never trust a statistic you didn’t fake with AI yourself.

[Franz und Franz @FranzUndFranz]

@Saanvi_dhillon Share the prompt :-D

[Franz und Franz @FranzUndFranz]

@benthompson Hihi - I am not using ultra anymore for the same reason. Btw check the file sizes of your codex session files.

[Franz und Franz @FranzUndFranz]

@petergostev Luckily we’re safe. It hasn’t managed to plant any seeds or settle in anywhere. Yet …

[Franz und Franz @FranzUndFranz]

@svpino A fool with a tool is still a fool…

[Franz und Franz @FranzUndFranz]

@ElitzaVasileva Nice punchline, but doing it the 2023 way in 2026 gives you far better results. Long prompts didn’t disappear. They became skills and instructions. Short prompts only work because you did the groundwork upfront. And even in 2026, no model is good at architecture.

[Franz und Franz @FranzUndFranz]

Don’t understand the urge to trash AI models. ChatGPT ≠ Claude ≠ Grok ≠ Gemini. Different tools, different strengths. Use accordingly. But no Chinese models for me - I don’t care how cheap they are.

[Franz und Franz @FranzUndFranz]

@psomkarbuilds 1 day coding means 2-3 days debugging. No matter what model. The better the model the more sophisticated is the debugging. So, yes, I read most of the code.

[Franz und Franz @FranzUndFranz]

@MatthewBerman @thsottiaux Btw - if you do so - check the file sizes of your session files in ~/.codex directory - maybe there is a „surprise“

[Franz und Franz @FranzUndFranz]

@rezoundous :-) :-) :-) I am not using ultra anymore for regular things - it overcomplicates everything. Btw - check the file sizes of your ~/.codex directory - session files

[Franz und Franz @FranzUndFranz]

@rezoundous If you are under 10%: a) you have coded too much, take a break! :-) Because b) you need 2x the codex time to fix all bugs codex introduced (or Claude) - 1-2 days coding - 3-4 days bug fixing.

[Franz und Franz @FranzUndFranz]

https://t.co/AIgLbp1aRC

[Franz und Franz @FranzUndFranz]

@rezoundous No

[Franz und Franz @FranzUndFranz]

@thsottiaux @Ananth7e Hopefully, you've set up your Codex Agent so that it keeps doing that for you...

[Franz und Franz @FranzUndFranz]

@thsottiaux 2 games, 3 tools, 2 website relaunches, and an AI memory system that’s going to revolutionize everything. And you made it all possible with the resets. Thank you so much!

[Franz und Franz @FranzUndFranz]

Ah, issue already reported here: https://t.co/WPurpXk5hn

[Franz und Franz @FranzUndFranz]

Hi Tibo - Codex CLI 0.144.5 still embeds the full parent rollout in each legacy-history subagent JSONL (GPT-5.6-sol, ultra). We see child files up to 118 MB; downstream importers can replay millions of messages. Intended? Can paginated/reference history become default? @thsottiaux

[Franz und Franz @FranzUndFranz]

@thsottiaux I love it without limits. Sol 5.6 rocks hard! Only big problem atm: if codex is running more than 24h without restart, there is a memory overflow. One codex goal was running for 72h and consumed 18gb ram.

[Anatoli Kopadze @AnatoliKopadze]

Anthropic engineer:

"Fable 5 is already smarter than we know how to use. The bottleneck was never the AI, it's you."

In 19 minutes he shows exactly how to get everything out of Claude with no extra tools, no extra costs.

You're already paying for all of this, just not using it.

Watch the session, then read the guide below on the Claude features 99% of users never find.

[Franz und Franz @FranzUndFranz]

@thsottiaux do the reset now!!! :-D tx!!!! gpt-5.6 is wonderful