#codingagents — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #codingagents, aggregated by home.social.
-
Your coding agent just spent 10 million tokens to fix a single bug. Is it worth it? 😉
Other researchers have found similar results. Bai et al. (2026) measured coding agents on real software tasks and found they consume roughly 1,000 times the tokens of an ordinary chatbot interaction. And these sort of agents tasks represent the most rapid driver of increased AI usage; Anthropic’s Economic Index found that 97% of their API usage now show “automation-dominant” patterns associated with agents.
#agents #codingagents #greensoftware #sustainability #ai #datascience #research #programming #python #opensource #GreenOps #tech #technology #data #technology #climatechange
https://www.theclimatebrink.com/p/the-real-energy-use-of-agentic-ai
-
Your coding agent just spent 10 million tokens to fix a single bug. Is it worth it? 😉
Other researchers have found similar results. Bai et al. (2026) measured coding agents on real software tasks and found they consume roughly 1,000 times the tokens of an ordinary chatbot interaction. And these sort of agents tasks represent the most rapid driver of increased AI usage; Anthropic’s Economic Index found that 97% of their API usage now show “automation-dominant” patterns associated with agents.
#agents #codingagents #greensoftware #sustainability #ai #datascience #research #programming #python #opensource #GreenOps #tech #technology #data #technology #climatechange
https://www.theclimatebrink.com/p/the-real-energy-use-of-agentic-ai
-
Fresh IBM Bob tasks can read the repository again, but they do not automatically recover why an earlier design was rejected.
This tutorial connects Engram's local knowledge graph to Bob 2.0 with project-scoped MCP. It also uses `.bob/rules/engram.md` for recall because Bob has no session-start hook.
I cover the Open VSX pane, stdio MCP process, HTTP/SSE daemon, shared SQLite graph, and a fresh-task recall test.
https://www.the-main-thread.com/p/ibm-bob-engram-durable-project-memory
-
Fresh IBM Bob tasks can read the repository again, but they do not automatically recover why an earlier design was rejected.
This tutorial connects Engram's local knowledge graph to Bob 2.0 with project-scoped MCP. It also uses `.bob/rules/engram.md` for recall because Bob has no session-start hook.
I cover the Open VSX pane, stdio MCP process, HTTP/SSE daemon, shared SQLite graph, and a fresh-task recall test.
https://www.the-main-thread.com/p/ibm-bob-engram-durable-project-memory
-
DeepSeek V4 Flash 0731 is interesting not because it became larger, but because it became much more capable through post-training.
The same architecture now delivers stronger coding-agent and tool-use performance while remaining remarkably inexpensive through the API.
https://open.substack.com/pub/doriandiaconu/p/deepseek-v4-flash-0731-the-ai-upgrade
-
DeepSeek V4 Flash 0731 is interesting not because it became larger, but because it became much more capable through post-training.
The same architecture now delivers stronger coding-agent and tool-use performance while remaining remarkably inexpensive through the API.
https://open.substack.com/pub/doriandiaconu/p/deepseek-v4-flash-0731-the-ai-upgrade
-
RT @0x0SojalSec: TRANSLASION: Z ai reportedly skipping GLM-5.3 and going straight to GLM-5.5 a leak 3T+ parameter open-weight model targeting Fable 5 / Mythos-level performance. - August release window - 1M token context - Built for coding, long-running agents - Still open-weight, 1M context - Targeting Fable 5 and Mythos-level capability - Focused on coding, reasoning, and autonomous agents
mehr auf Arint.info
#AIModel #CodingAgents #GLM55 #MythosLevel #OpenWeight #Zai #arint_info
-
RT @0x0SojalSec: Z ai reportedly überspringt GLM-5.3 und geht direkt zu GLM-5.5, einem Leak mit 3B+ Parametern als open-weight Modell, das Fable 5 / Mythos-Level Performance anstrebt. - Veröffentlichungsfenster im August - 1M Token Kontext - Entwickelt für Coding, langlaufende Agenten - Immer noch open-weight, 1M Kontext - Ziel: Fable 5 und Mythos-Level Fähigkeit - Fokus auf Coding, Reasoning und autonome Agenten
mehr auf Arint.info
#AIModels #CodingAgents #GLM55 #MachineLearning #OpenWeight #Zai #arint_info
-
Just hit 100 🌟 on GitHub for LoopTroop v0.4!
LoopTroop is a free & open-source local AI coding orchestrator that turns tickets into real PRs without endless chat drift.
⚙️ LLM councils plan the work
🔁 Ralph loops iterate & perfect it
🌲 OpenCode worktrees ship the PR
Built with human approval gates, planned task beads, and local control.
The first 100 stars are the hardest. Huge thanks to everyone supporting the project!
Clone it, run npm run dev, and try it on your repo:
📦 Repo: https://github.com/looptroop-ai/LoopTroop
🌐 Site: https://looptroop.ovh#LoopTroop #opensource #AgenticAI #BuildInPublic #AI #DevTools #CodingAgents
-
ELIZA is a chatbot written in the 1960s. It’s very relevant in 2026.
-
ForkMesh makes it ridiculously easy to manage multiple coding agents at once. Each agent works on its own issue, worktree, and branch, keeping tasks isolated while the coding pipeline stays fast, parallel, and efficient. Launch more agents, avoid conflicts, and ship faster. 🚀 https://forkmesh.com/r/jett
#ForkMesh #CodingAgents #AgenticAI #SoftwareDevelopment #DevTools #DeveloperTools #Programming #Coding #Git #GitBranches #GitWorktrees #Automation #DevAutomation #Engineering #SoftwareEngineering
-
Why Software Factories Fail (or: harness engineering is not enough)
https://github.com/humanlayer/advanced-context-engineering-for-coding-agents/blob/main/wsff.md
Comments: https://news.ycombinator.com/item?id=49023019
#HackerNews #softwarefactories #softwareengineering #techfailures #engineeringhacks #codingagents
