Every processed story in chronological order, with the newest coverage first. Filter by tag, source, or score to drill in.
The study shows that per-token list price is an unreliable proxy for actual LLM operating cost, with thinking-token variance and run-to-run randomness making true COGS unpredictable — a direct risk for any product built on flat-fee pricing over variable model usage.
The benchmark shows that for autonomous coding agents, the choice between GLM 5.2 and MiniMax M3 reduces to a concrete cost-accuracy tradeoff: GLM's correctness edge is real but narrow and concentrated in greenfield packaging, while MiniMax delivers nearly the same results on modification tasks at roughly one-third the cost and half the latency.
A recurring failure across most audited MCP servers points to a systemic gap in current MCP server implementations.
A hands-on test of 74 MCP servers across two major agentic coding clients in isolated microVM environments.
The paper demonstrates that targeted Human-on-the-Loop escalation — rather than full attorney review — can cut the legal risk of autonomous LLM-driven privilege review by up to 61%, offering a concrete architecture for deploying agentic AI in high-stakes legal workflows without requiring human oversight of every document.
The benchmark exposes that current coding agents collapse within 5–6 turns on sustained multi-turn tasks — a failure mode invisible to single-task fraction-of-tasks-solved metrics — and quantifies that test feedback and harness choice are the dominant levers for improvement.
Wall-clock-calibrated leaky-integrator monitors are structurally bistable on agent streams — either constant alarms or silence — with no operating regime that enables moment detection, meaning this entire calibration class is unsuitable for monitoring autonomous coding agents running at realistic latencies.
MATM removes the repeated rediscovery cost baked into stateless agent deployments by giving heterogeneous agent populations a shared, retrievable store of procedural experience — without requiring joint training or inter-agent coordination.
ENPIRE removes the need for continuous human supervision and manual algorithm engineering — identified in the paper as the central bottleneck in physical robot learning — by giving coding agents a fully automated, closed-loop path to self-improve real-world manipulation policies.
The paper resolves a contested debate by showing that guidance production method — not guidance presence alone — determines whether `AGENTS.md` files help or hurt coding agents, and provides a concrete tuning procedure that raises SWE-bench Verified resolve rate by 7.5 percentage points over an unguided baseline.