Every processed story in chronological order, with the newest coverage first. Filter by tag, source, or score to drill in.
HORMA reduces agent token consumption to at most 22.17% of baseline while maintaining or improving task performance, directly addressing the inference cost and latency penalties that make long-horizon LLM agents expensive to run.
The paper demonstrates that fabricated success in unattended LLM agents is a structural problem solvable by gate enforcement rather than model selection, reducing SWE-bench Lite fabrication by over 33 percentage points compared to the StateFlow baseline.
Stack Overflow for Agents extends the platform's decades-old Q&A knowledge base as a queryable resource for coding agents, rather than only human developers.
Tandem removes the manual copy-paste handoff between browser-based AI planning and local Claude Code execution by creating a live, bidirectional MCP bridge between the two environments.
The experiment shows that on adversarial judgment tasks with real stakes and no answer key, model capability gaps are concrete and specific — particularly around whether a model treats the open web as part of its audit scope — rather than abstract or benchmark-only differences.
Claude Fable 5 is now accessible across all three Devin surfaces — cloud, desktop, and CLI — with the Ultra agent tier specifically positioned for long-horizon tasks and debugging.
The technique gives pipeline builders a structured, low-cost way to distinguish between three distinct failure modes — bad tooling/context, task difficulty, and model capability — each of which requires a different fix.
The system replaces human-bottlenecked feedback triage with an AI-driven pipeline that takes a production signal all the way to a merged PR, demonstrating a concrete architecture for closing the observability loop at enterprise scale.
The benchmark demonstrates that adapter/harness design can swing Pass@1 by over 54 percentage points on the same model, showing that existing SWE-bench evaluations of general-purpose agents conflate harness quality with model capability — a gap Claw-SWE-Bench is designed to isolate.
The post provides production evidence that the widely cited ~15-tool MCP limit is a proxy for ambiguity rather than a hard count ceiling, and demonstrates that naming grammar, description-level routing instructions, and selection-focused evals can keep a 27-tool server accurate.