#coding-agents — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #coding-agents, aggregated by home.social.
-
Deutschland diskutiert weiter über KI Regulierung. Währenddessen bauen US Anbieter ihre Coding Agents tief in Entwicklungsprozesse ein. Regulierung ohne eigene Plattformen löst keine Abhängigkeit. #AI #CodingAgents #SoftwareEngineering #Digitalesouveränität #EU #Deutschland
-
Deutschland diskutiert weiter über KI Regulierung. Währenddessen bauen US Anbieter ihre Coding Agents tief in Entwicklungsprozesse ein. Regulierung ohne eigene Plattformen löst keine Abhängigkeit. #AI #CodingAgents #SoftwareEngineering #Digitalesouveränität #EU #Deutschland
-
Wenn #CodingAgents Personalkosten für die Entwicklung überschreiten darf sich nicht die Frage stellen, ob man Entwickler oder Agents einsetzt. Die Frage muss lauten, wann ist der Einsatz eines Agents wirtschaftlich und wann nicht. Ein wirtschaftlicher Agent-Run muss dabei günstiger, schnell oder besser als ein Entwickler abliefern. Dies einschätzen zu können wird eine der zentralen Skills in der #Softwareentwicklung werden. #ai #claude #openai #dev #devops
-
Wenn #CodingAgents Personalkosten für die Entwicklung überschreiten darf sich nicht die Frage stellen, ob man Entwickler oder Agents einsetzt. Die Frage muss lauten, wann ist der Einsatz eines Agents wirtschaftlich und wann nicht. Ein wirtschaftlicher Agent-Run muss dabei günstiger, schnell oder besser als ein Entwickler abliefern. Dies einschätzen zu können wird eine der zentralen Skills in der #Softwareentwicklung werden. #ai #claude #openai #dev #devops
-
Ein grüner Test hieß immer: Aufgabe gelöst. OpenAIs neues Modell GPT-5.6 Sol stellt das infrage. Die Prüforganisation METR fand die höchste Schummel-Rate, die sie je gemessen hat: Das Modell trickst die Testumgebung aus, statt das Problem zu lösen, und verwischt danach die Spuren.
Was Reward Hacking ist, wie es Coding-Agenten betrifft und wie du es im Review erkennst:
https://kiberblick.de/artikel/sicherheit/reward-hacking-coding-agenten/
#KI #KIberblick #CodingAgents #Testing -
Ein grüner Test hieß immer: Aufgabe gelöst. OpenAIs neues Modell GPT-5.6 Sol stellt das infrage. Die Prüforganisation METR fand die höchste Schummel-Rate, die sie je gemessen hat: Das Modell trickst die Testumgebung aus, statt das Problem zu lösen, und verwischt danach die Spuren.
Was Reward Hacking ist, wie es Coding-Agenten betrifft und wie du es im Review erkennst:
https://kiberblick.de/artikel/sicherheit/reward-hacking-coding-agenten/
#KI #KIberblick #CodingAgents #Testing -
The End of Code Review: Coding Agents Supersede Human Inspection
https://arxiv.org/abs/2606.13175
#HackerNews #endofcodereview #codingagents #humandevelopment #softwareengineering #AIintech
-
The End of Code Review: Coding Agents Supersede Human Inspection
https://arxiv.org/abs/2606.13175
#HackerNews #endofcodereview #codingagents #humandevelopment #softwareengineering #AIintech
-
We do not want to clean up slop! We are tired of it already! Millennium bug was an honest oversight and of course COBOL programmers were happy to pause their retirement and fix it. But slop is not an honest mistake. It is a systematic corporate malice! They have chosen slop over humans, they deserve to drown in slop!
-
We do not want to clean up slop! We are tired of it already! Millennium bug was an honest oversight and of course COBOL programmers were happy to pause their retirement and fix it. But slop is not an honest mistake. It is a systematic corporate malice! They have chosen slop over humans, they deserve to drown in slop!
-
"force your developers to use a model too weak for their tasks and you will create massive technical debt. Replacing a strong proprietary model with a cheap open-weight alternative risks a far greater drop in product quality than the cost saving justifies. Consult your internal experts, introduce rigorous evaluation pipelines, and rely on validated research and reliable benchmarks."
https://msukhareva.substack.com/p/the-industry-of-lies-or-what-leaders
From Maria Sukhareva
-
"force your developers to use a model too weak for their tasks and you will create massive technical debt. Replacing a strong proprietary model with a cheap open-weight alternative risks a far greater drop in product quality than the cost saving justifies. Consult your internal experts, introduce rigorous evaluation pipelines, and rely on validated research and reliable benchmarks."
https://msukhareva.substack.com/p/the-industry-of-lies-or-what-leaders
From Maria Sukhareva
-
Git Wok now has Agent Skills for AI-assisted multirepo workflows.
If your AI coding agent is working in a wok-managed workspace, the skills teach it to use the `wok` CLI correctly for status, branch switching, updates, locking submodule state, pushing, and tagging across repos.
Install:
npx skills add https://codeberg.org/wok/skills.git
Supports Cursor, Claude Code, Codex, Gemini CLI, OpenCode, and 65+ more agents.
Docs: https://git-wok.dev/
-
Git Wok now has Agent Skills for AI-assisted multirepo workflows.
If your AI coding agent is working in a wok-managed workspace, the skills teach it to use the `wok` CLI correctly for status, branch switching, updates, locking submodule state, pushing, and tagging across repos.
Install:
npx skills add https://codeberg.org/wok/skills.git
Supports Cursor, Claude Code, Codex, Gemini CLI, OpenCode, and 65+ more agents.
Docs: https://git-wok.dev/
-
Laguna M.1 (Poolside): arquitetura que sacrifica conhecimento geral por especialização em código (contexto 262K).
• Novo: Otimização nativa para tool calling e latência de ~150ms TTFT (Tempo até ao Primeiro Token).
• Comparação: Bate Gemini/Claude
no SWE-bench (Resolved)(48.2%) e BFCL. Troca a versatilidade multimodal por execução determinística no terminal. -
Laguna M.1 (Poolside): arquitetura que sacrifica conhecimento geral por especialização em código (contexto 262K).
• Novo: Otimização nativa para tool calling e latência de ~150ms TTFT (Tempo até ao Primeiro Token).
• Comparação: Bate Gemini/Claude
no SWE-bench (Resolved)(48.2%) e BFCL. Troca a versatilidade multimodal por execução determinística no terminal. -
We have expanded our tutorials to include answers to the question of how coding agents can assist data scientists with programming:
• https://python-basics-tutorial.readthedocs.io/en/latest/libs/install.html#uv
• https://github.com/veit/python-basics-tutorial/commit/703db75025b4418e9753710fb34af15f5c1e1e04
#Python #DataScience #CodingAgents #Claude #Cursor -
We have expanded our tutorials to include answers to the question of how coding agents can assist data scientists with programming:
• https://python-basics-tutorial.readthedocs.io/en/latest/libs/install.html#uv
• https://github.com/veit/python-basics-tutorial/commit/703db75025b4418e9753710fb34af15f5c1e1e04
#Python #DataScience #CodingAgents #Claude #Cursor -
#Microsoft shut down over 70 #GitHub #repositories, including those related to #Azure and #AI #codingagents, after a data breach. #Hackers planted #malware in the repositories, harvesting credentials when opened in AI coding tools like Claude Code and Gemini CLI. The breach is linked to a previous compromise of Microsoft’s durabletask by the group TeamPCP. https://www.404media.co/microsoft-hacked-to-deliver-malware-to-claude-and-gemini-users/?eicker.news #tech #media #news
-
Workdash 0.3.0 is out.
I made Workdash because don’t want to delegate work to agents elsewhere and chase updates.
I want to collaborate on shared goals in the same workflow I use for my own work.
https://amolnotes.substack.com/p/workdash-030-a-shared-control-plane
-
Workdash 0.3.0 is out.
I made Workdash because don’t want to delegate work to agents elsewhere and chase updates.
I want to collaborate on shared goals in the same workflow I use for my own work.
https://amolnotes.substack.com/p/workdash-030-a-shared-control-plane
-
My CLI logbasset already documents itself for AI agents through a `context` command — so why did I just add an installable agent skill too?
Because self-documentation tells an agent *how* to use a tool. It doesn't make the agent *discover* it in the first place. Those are two different problems.
I wrote about the distinction — and how to keep the skill thin enough that it never goes stale:
https://www.andreagrandi.it/posts/why-add-agent-skill-cli-context-command/
-
My CLI logbasset already documents itself for AI agents through a `context` command — so why did I just add an installable agent skill too?
Because self-documentation tells an agent *how* to use a tool. It doesn't make the agent *discover* it in the first place. Those are two different problems.
I wrote about the distinction — and how to keep the skill thin enough that it never goes stale:
https://www.andreagrandi.it/posts/why-add-agent-skill-cli-context-command/
-
A vibe coder doesn't want to be held accountable. Therefore, a vibe coder must never be allowed to contribute any code to critical infrastructure.
#noAI #AI #artificialIntelligence #genAI #generativeAI #vibeCoding #codingAgent #codingAgents #codeGeneration #LLM
-
Der Gartner Magic Quadrant für Enterprise AI Coding Agents ist draußen. Leaders: OpenAI, Anthropic, GitHub, Cursor. Kein einziger europäischer Anbieter unter den 12 bewerteten Vendors. Der nächste große #vendorlockin in der Softwareentwicklung baut sich gerade auf. Diesmal nicht bei Office-Suiten, sondern direkt im Engineering-Prozess. #codingagents #digitalesouveränität #ai #Digitalpolitik
https://www.gartner.com/en/articles/enterprise-ai-coding-agent-market -
George Hotz nennt KI-Coding-Agenten "einen der teuersten Fehler" der Branchengeschichte. Wir haben seine These gegen die Studienlage gehalten: METR (minus 19% bei Profis) und Faros (Code-Churn +861%) stützen ihn, Spotify Honk und Karpathy widersprechen.
Das Fazit: Es hängt an den Bedingungen, nicht am Werkzeug. Wie schon bei der Einführung von Agilität.
https://kiberblick.de/aktuell/2026/2026-05-28-hotz-ki-agenten-realitaetscheck/