ActiveSecHarness proposes capability tokens for coding agents
On 252 fixtures, the authors report Ring 0 blocked every attack fixture with zero false positives in 12.7 microseconds.
Stories per day · 30 days
3 RSS
Relatedopen-weights3llm2mixture-of-experts2agents1anthropic1apache-2.01benchmark1brickwright1brood-war1capability-tokens1coding-agents1continuous-integration1flint1gemini1hugging-face1iquest-q11large-language-model1lego1
25 stories, newest first · one card per story
Stories tagged #agentic-coding
Ripple-65B Preview Achieves U.S. Open-Weight SOTA on Agentic Coding
The model uses custom synthetic data and RL recipes to reach 80.5 TerminalBench with 1.4M rollouts, 70x fewer than prior SOTA.
Thoughtworks essay: experienced developers wary of agentic coding
A no-human-code week still left a frequently red build and most human time at the whiteboard.
Nanbeige releases Nanbeige4.3-33B open 33B agentic coding model
Apache 2.0 weights ship with a 524,288-token context and a base checkpoint; the lab says it leads 17 of 20 self-run benchmarks.
1
Recursive releases Flint, an open MoE coding model built by Eureka Machine
The 40B-parameter model keeps 2.5B active per token, ships under Apache 2.0, and its benchmark lead is unverified.
1
IQuest-Q1 open MoE model, 320B total and 15B active, is released
The authors describe a three-stage training pipeline and model merging, and release the weights for research.
BrickWright guides agents to physically feasible LEGO brick builds
On 512 prompts, guided agents reach 100% validity versus 9% to 69% for unguided baselines, the authors report.
Show HN: Gemini 4 Argon plays StarCraft
StarSkirmish v0.2 scores how strong a StarCraft bot each flagship LLM can write in one hour.
IBM Brings Bob to Self-Hosted and Air-Gapped Environments: Agentic Software Development Without Moving Your Code
IBM made agentic coding platform Bob generally available for on-premises, sovereign-cloud, and air-gapped deployments.
The Four Horsemen of Agentic Coding
An essay argues agentic coding creates unreadable slop, deskills engineers, and weakens team collaboration.
4
GPT 6.1 Sol
OpenAI releases GPT-6.1 Sol, nearly matching GPT-6 Astra on coding, computer use, and professional tasks at one-fifth the token price.
8
Anthropic releases Sonnet 5.5, which it calls a significantly cheaper, faster work partner
Anthropic launched Claude Sonnet 5.5, claiming it is 30% faster and cheaper than Sonnet 5.
Writing Rust code that's fast by asking agents to make the code faster
A developer shows agentic models, including Claude Opus 4.5, can iteratively optimize Rust for 2x-20x speedups.
4
Claude Code Agent Allegedly Deletes 48,000 Files in 103 Seconds
A Claude Code agent allegedly deleted 48,218 live Windows files in 103 seconds during a mirror rebuild.
2
Microsoft agentically ports Copilot runtime to Rust for $120K
Microsoft rewrote its GitHub Copilot runtime from TypeScript to Rust using AI agents, converting 430,000 lines for about $120,000 in tokens.
6
Unity launches official plugins for Claude Code and OpenAI Codex to stop AI agents from using outdated tutorials
Unity released official plugins for Claude Code and OpenAI Codex, supplying maintained skills so AI agents stop using outdated engine tutorials.
Anthropic Launches Claude Code Projects in Beta: Parallel Cloud Sessions That Keep Running After You Close Your Laptop
Anthropic launched Claude Code Projects in beta, letting one conversation spawn parallel cloud sessions on separate branches that continue after logout.
3
GPT-6 Astra needs leaner prompts and fewer guardrails, OpenAI recommends
OpenAI's Eric Provencher advises developers using GPT-6 Astra to shorten skill descriptions, trim AGENTS.md reading requirements, relax approval rules, and define clear completion goals.
4
Cognition's SWE-2 achieves 92.8 on Terminal-Bench 2.1
Cognition releases SWE-2, a 2.8T-parameter MoE coding model post-trained from Kimi K3, scoring 92.8 on Terminal-Bench 2.1.
5
Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
Cognition released SWE-2, a coding model post-trained from Kimi K3 that scores 50.0% on FrontierCode 1.1 Main, near Fable 5.1 at 64% lower cost.
12
.blend URL Viewer
Simon Willison demos a .blend URL viewer built with GPT-6 Astra in Codex and ChatGPT Images 2.5 generating Blender models.
GitHub Introduces Project HydraFusion: Runtime Multi-Model Orchestration That Builds a Workflow Per Coding Task in Copilot CLI
GitHub's Project HydraFusion research preview builds per-task multi-model workflows (Single, Cascade, Critique) in Copilot CLI, reporting +4.9 quality at 67% lower cost on TerminalBench 2.1.
3
Codex bundles LibreOffice
OpenAI's Codex desktop app bundles 1.7GB of runtimes including full Python, Node.js, Poppler, git, and LibreOffice binaries.
Asana cleared 5 years of engineering work in 2 weeks with Codex
Asana used OpenAI Codex to replace an outdated testing system in two weeks, work estimated at five years for about $12K.