Every processed story in chronological order, with the newest coverage first. Filter by tag, source, or score to drill in.
The study reveals that mainstream benchmarks like SWE-Bench Verified and Terminal-Bench 2.0 compress capability differences between agents into narrow bands, and that esoteric language evaluation exposes a qualitative gap in how strong versus weak agents construct and debug novel strategies.
Role-Agent demonstrates that a single LLM can bootstrap its own agent training by self-generating both process rewards and targeted practice tasks, achieving consistent gains over strong baselines without requiring separate environment models.
Lore addresses a concrete, largely silent failure mode in long-running AI coding sessions — context compaction — by replacing it with a persistent, searchable memory pipeline that works across sessions, tools, and team members without requiring workflow changes.
The architecture demonstrates that constraining LLM involvement to structured front-end parsing — rather than solver code generation — can achieve high reliability on finite element simulation benchmarks while avoiding the code-correctness risks of open-ended autonomous generation.
Red Queen addresses a gap the source identifies — the lack of a deterministic, auditable pipeline layer above existing AI coding agents — by providing token-free routing, configurable human gates, and retry-with-escalation logic as first-class workflow primitives.
This project fills the gap left by the absence of an official Anthropic Claude Desktop release for Linux, providing native packages across the major Linux distribution families.
The `oneOf`/`allOf` schema preservation and web search additions extend Codex's compatibility with richer MCP tools and broaden what code mode agents can do without leaving the coding workflow.
The release pairs access to Anthropic's most capable generally available model to date with a fix for a session-continuity regression that affected VS Code integrated terminal users.
The feature gives teams a concrete guardrail against runaway AI spend, particularly for autonomous or unsupervised workflows that can consume tokens faster than manual monitoring can catch.
The study establishes automated prompt injection as a credible but model-dependent threat to LLM agents, while identifying significant barriers — particularly the failure of smaller-model attacks to transfer to frontier models — that shape the realistic risk landscape for agentic systems.