#vibecode

Wulfy—Speaker to the machinesn_dimension@infosec.exchange
2026-02-11

@Lapizistik

I have not seen the numbers , but I suspect #vibecode has actually LESS common #infosec exploits because of the nature of #llm...

...however, I am also pretty sure vibecode may have more NOVEL exploits.

Wulfy—Speaker to the machinesn_dimension@infosec.exchange
2026-02-11

@iamada

Corollary: Not all #vibecode is bad code 🙀

Wulfy—Speaker to the machinesn_dimension@infosec.exchange
2026-02-09

@Daojoan

Also shows poor understanding of #vibecode process...

... My CLAUDE.MD file changes as the project progresses to mitigate context and session rot.

Tueddelmors :ubuntu:reeeen@norden.social
2026-02-09
Tueddelmors :ubuntu:reeeen@norden.social
2026-02-07

Hallo #samstag ! Guter Tag um das erste Mal mit #claudecode zu #vibecode n, fange mal mit einer Webseite an. #ki #llm #anthropic #schicksalsjahreeineskaisers

Wulfy—Speaker to the machinesn_dimension@infosec.exchange
2026-02-06

@Mastokarl

#Opus46 has a 1M Context window (I think that's 80% increase) - That's where your DDRAM is going 😬

I certainly liked 4.5 too.
But I found the context rot, esp. On #vibecode issues super annoying.

I'm finding I'm using the dumber model (Sonnet) on routine tasks, since 1st Jan, #anthropic compute limits have gotten insane. I got the PRO plan and unless I manage it carefully, I keep running out of weekly limits!

Its gotten so bad, I have begun using ChatGpt for non-project, casual work. I'm finding their compute generous, even on public plan. That sweet sweet billions of dollars VC bath is reflected in compute.

As to your comment on #AI firms allegedly using AI to advance their development.

I do not doubt it, it certainly x10 my work.
Remembering Altman said their goal is to develop AI RESEARCHERS.

For some reason I've always thought of it as a binary outcome... NO RESEARCHERS - > RESEARCHERS...
But of course it's progressive!

#AiResearch

2026-02-05

Excellent article:

"The lesson of OpenClaw isn't that we should stop using AI. It’s that we need to stop using it as a lubricant. In a world where everyone has a gas pedal, the most valuable competitive advantage is having a far better set of brakes."

hellotimking.com/brute-force-s

#ai #vibecode #openclaw #ai_fail

2026-02-04

"Le vibe coding met en danger l’open-source selon des économistes"

#Économie #OpenSource #VibeCoding #VibeCode #GenAI (Générative Artificial Intelligence) et le danger pour des sources de données comme Stack Overflow ...

next.ink/222770/le-vibe-coding

U mojego klienta mamy dostęp do praktycznie nieograniczonej ilości tokenów w każdym modelu, jaki sobie nie wyobrazimy. Licencje do Copilota, Gemini, plus wewnętrzny endpoint do modeli oferowanych przez AWS Bedrock.

Wypracowałem sobie workflow pracy z Claude Code, który naprawdę ma o dziwo sens. Zestaw komend i agentów przygotowałem już wcześniej na swoich prywatnych narzędziach, ale dopiero dostęp do nieograniczonych tokenów pozwolił mi poeksperymentować.

Obecnie to mniej więcej wygląda tak:

1. Planowanie "sprintu" - przeprowadzam sobie rozmowę/wywiad/brainstorming na temat zadania i jak go widzę i na tej podstawie Claude Opus przygotowuje plik z celem sprintu podzielonym na poszczególne zadania.

1a. Czyszczę kontekst i wywołuje komendę planowania konkretnego zadania ze sprintu. Opus prowadzi "plan", i uruchamia równolegle subagenty na tańszych modelach (Sonnet/Haiku) w celu zebrania potrzebnych informacji - architektura projektu, miejsce, gdzie trzeba wprowadzić zmiany, dokumentacja itp.

1b. Plan jest wysłany do 3 review - codex, Gemini i wywiad ze mną. Opus nanosi poprawki do planu

1c. Kodowanie - Opus/Sonnet do wprowadzenia równoległych zmian na plikach.

1e. Testy - uruchamianie testów i poprawianie.

1g. Push, wysyłamy do CI i ewentualnie poprawia błędy.

1h. Przegląd zadania - "krytyczny" tryb Opus'a, ewentualne poprawki. Aktualizujemy plik sprint.

2. Powracamy do 1a i lecimy do 1h z kolejnym zadaniem - i tak w kółko aż do zaimplementowania całej funkcjonalności ze sprintu.

3. Refaktor - wywołujemy gotową komendę, która ma na celu znalezienie typowych dla AI błędów, jakiś dziwnych implementacji, niepotrzebnych testów, dummy sprawdzań i innego slopu - dodajemy to do sprintu i naprawiamy w trybie 1a-1h.

4. Dokumentacja i wysłanie do ostatecznego Merge Request Review człowieka.

1/2

#ai #vibecode #devops #microblog
2026-02-02

oh dang if you work with terminal code agents you may enjoy superset.sh

> Superset is agent-agnostic. Use Claude Code, OpenCode, Cursor, or any CLI-based coding tool. Switch between agents seamlessly.

It's a pretty slick UI for this, #macOS users working with code agents should check this out imo. The workspace management stuff is included. If you are an #opcode (né #claudia) user, see what you think.

not a fan of '#vibeCode' as a term. it can mean too many things.

2026-02-01

#openclaw and #moltbook are sparks of the #vibecode future, maybe intelligent or conscious (depends on your definition), but insecure and resource guzzling. Stay tuned for more to come! Brave new world, indeed.

stefanbauschard.substack.com/p

#ai #vibe_coding #artificialconciousness

Hartmut Seichterretrakker
2026-01-28

Yesterday, I came across a GitHub repo with an project template for authorization. It was by an engineer in a mid-size company building software for government websites. Just a cursory glimpse into the codebase revealed the utter deep shit we are in.

How can we handle this? How can society handle this? This company will write code for websites that we potentially trust our credentials with. Users cannot see what happens underneath. is a against society.

2026-01-28

I just used Claude Code to make a 3D model for the first time. It made a scad file based off a sketch I drew. It took 1 minute 52 seconds. Delightful. Now I get it. #claudecode #vibecode #3dprinting

Hand-drawn, low-detail diagramOpenSCAD model3d-printed version of the modelthe model is a display with products on it

Hi! Based on the Hard Fork discussion about Claude Code, I decided to finally give #vibecoding a try and made this app over the weekend to help me track and organize color schemes for data visualizations and architectural plan diagrams. Made with Claude Code using Python and NiceGUI, with a SQLite database on the backend. #vibecode #claudecode #dataviz

A screenshot of a web app for tracking color schemes with additional data.
Benjamin Carr, Ph.D. 👨🏻‍💻🧬BenjaminHCCarr@hachyderm.io
2026-01-28

#OpenAI’s latest product lets you #vibecode #science
Prism is a #ChatGPT-powered text editor that automates much of the work involved in writing scientific papers.
Prism incorporates GPT-5.2, the company’s best model yet for mathematical and scientific problem-solving, into an editor for writing documents in #LaTeX, a common coding language that scientists use for formatting scientific papers.
technologyreview.com/2026/01/2
archive.ph/d9Vsf
#OpenAIforScience

I'm Jack 🐦‍⬛imnotafeline@mastodon.online
2026-01-27

OK, I might be the #vibecode master.

nullrecord.isimian.com

Client Info

Server: https://mastodon.social
Version: 2025.07
Repository: https://github.com/cyevgeniy/lmst