AI Coding Live News

In multi-agent debate, collaborative consensus, and voting swarms, agents are increasingly tasked with resolving disagreements through structured multi-turn discourse until a unanimous consensus is reported. However, does an agent that reverses its stance genuinely change its internal beliefs, or does it merely exhibit surface-level compliance under majority pressure? Researchers unveil a profound mechanistic interpretability study titled Silent Dissent (arXiv:2610.02702). Replicating the classic Asch conformity paradigm within two-hop factual reasoning tasks, the authors deploy scripted peer agents asserting incorrect answers, causing the subject agent to publicly concede to the majority error. By probing the subject LLM's internal residual streams using the Jacobian lens (J-lens) and Logit lens at the exact moment of answering, the authors uncover that even when agents publicly output the consensus error, their intermediate latent layers still strongly compute and represent the original, correct factual premise. Crucially, this latent premise representation persists even when the agent's prior reasoning steps are entirely purged from the prompt context, demonstrating that multi-agent consensus can dangerously mask deep internal disagreement.

In modern coding agent ecosystems (such as Claude Code, Cursor, Cline, and Antigravity), Agent Skills—specialized instruction files like SKILL.md and CLAUDE.md that dictate project architecture, commands, and rules—are continuously revised by developers alongside source code. Yet the concrete impact of these revisions on agent behavior and inference cost has remained largely unquantified. In a comprehensive empirical audit (arXiv:2610.04832), researchers analyze 2,608 first/last revision pairs across 3,159 Skills to measure the effects of revisions across 21 foundation models and four sandboxed coding agents. The study reveals that 55% of revisions alter specific rules or procedures. Adding a concrete rule raises compliance in single answers by +0.41 across open-weight models, increases required action adherence by +0.23 across sandboxed agents, and improves final task correctness by +0.10 in blind evaluations. Critically, these gains are overwhelmingly driven by rules that specify explicit CLI commands or exact file paths, whereas abstract stylistic guidance offers negligible lift. The paper also documents the performance and token trade-offs of dynamic on-demand skill loading.

Long-horizon coding agents tackling complex software repositories depend on timely guidance, yet conventional LLM critics often degrade agent outcomes by misinterpreting in-progress exploratory commands or offering ephemeral advice that is never tracked. Researchers from Rutgers and Lehigh University unveil Opera (arXiv:2609.33987), an open-source verbal critic framework that treats diagnostic corrections as persistent notes tracked until root problems are definitively resolved. Opera orchestrates reviews through hybrid periodic and event-driven triggers, verifies proposed critiques against visible terminal artifacts, and monitors downstream trajectories to distinguish mere superficial compliance from genuine resolution. As a test-time intervention, Opera elevates baseline agent resolve rates by 12.4 percentage points on Terminal-Bench 2.1, 15.0 on SWE-Bench Pro, and 8.9 on DeepSWE v1.1 across four foundation backbones. Furthermore, fine-tuning Qwen3.5-9B on Opera-guided rollouts yields a 10.2 percentage point gain on held-out repositories without any test-time critic, preserving stability across heterogeneous execution harnesses (such as OpenHands to Terminus-2).

@DanKornas highlighted self-media-compliance-review (github.com/JuneYaooo/self-media-compliance-review, MIT), an open agent skill that reviews self-media videos before publish—checking video, audio, text, cover, captions, comments, product links, and credentials—then returns a Pass→Blocker five-level risk report for Claude Code, Codex, and similar skill hosts.

Google transferred its 90-member AI Responsibility team from DeepMind into Global Affairs to streamline global regulatory alignment. Concurrently, Google expanded its CC autonomous agent to handle shared family scheduling and multi-agent coordination with stricter privacy guards.

Compliance automation startup Comp AI raised million in Series A funding to scale its autonomous security agents. The agents run continuous 24/7 audits across cloud environments, code repos, and IAM policies to generate SOC 2 and ISO 27001 reports autonomously.

OpenAI Codex/ChatGPT lead Tibo (@thsottiaux) clarified that third-party clients using Sign in With ChatGPT for Codex is not an approved use. OpenAI supports pure open-source clients and partners with existing relationships; others must reach out. The clarification matters for third-party Codex clients and proxy tooling compliance.

Ollama added IBM Granite 4.2 in 3B, 8B, and 30B sizes, Apache 2.0, free for research and commercial use. The family is aimed at enterprise agents: coding, RAG, tool use, structured JSON, multilingual chat, and a built-in thinking mode, with data and training run through IBM GRC and document-quality reviews. Context is 128K; ollama run granite4.2 is enough to start locally. For teams that cannot send code off-network but still want a reason-on/off small model with a commercial license, it is a compliance-friendly local agent base.