ResearchMETR·18d ago·3 min read
Metrics of Agent AbilityMETR published research on metrics for measuring agent ability.
agent
News Hub
What actually shipped in agent engineering, pulled from the labs, arXiv and Hacker News.
See who we follow →51 articles
ResearchMETR·20d ago·3 min read
The Economics of Recursive Self-ImprovementMETR published analysis on the economics of recursive self-improvement.
self-improv
ResearchMETR·21d ago·3 min read
Expenditure Horizon: Measuring Optimization Ability, with an Application to NanoGPTMETR published research on measuring optimization ability using NanoGPT as an application.
gpt
ResearchMETR·34d ago·3 min read
Because 8 ≈ e², Anthropic's researcher uplift is plausibly >2xTitle alone insufficient for concrete summary.
ResearchMETR·46d ago·3 min read
Summary of METR's predeployment evaluation of GPT-5.6 SolMETR published predeployment evaluation results for GPT-5.6 Sol.
evalgpt
ResearchMETR·84d ago·3 min read
Frontier Risk Report (February to March 2026)METR published a frontier risk report covering February to March 2026.
ResearchMETR·92d ago·3 min read
Measuring the Self-Reported Impact of Early-2026 AI on Technical Worker ProductivityMETR measured the self-reported impact of early-2026 AI on technical worker productivity.
ResearchMETR·95d ago·3 min read
Review of the "Risks from automated R&D" section in the Anthropic Risk Report (February 2026)METR reviews the automated R&D risks section from Anthropic's February 2026 Risk Report.
ResearchMETR·112d ago·3 min read
Evidence on AI R&D Progress from NanoGPTMETR studied AI R&D progress using NanoGPT as evidence.
gpt
ResearchMETR·132d ago·3 min read
Fine-tuning experiments on CoT controllabilityFine-tuning experiments investigate chain-of-thought controllability.
fine-tun
SafetyMETR·138d ago·3 min read
Red-Teaming Anthropic's Internal Agent Monitoring SystemsMETR red-teamed Anthropic's internal agent monitoring systems.
agent
ResearchMETR·152d ago·3 min read
Review of the Anthropic Sabotage Risk Report: Claude Opus 4.6METR reviews Anthropic's sabotage risk report for Claude Opus 4.6.
claude
ResearchMETR·161d ago·3 min read
Observations from two CLI game reimplementation runs with Opus 4.6METR observed results from two CLI game reimplementation runs using Anthropic's Opus 4.6 model.
ResearchMETR·175d ago·3 min read
Analyzing coding agent transcripts to upper bound productivity gains from AI agentsResearchers analyzed coding agent transcripts to estimate upper bounds on productivity gains from AI agents.
agent
ResearchMETR·179d ago·3 min read
Measuring Time Horizon using Claude Code and CodexMETR measured time horizon using Claude Code and Codex.
claude
ResearchMETR·201d ago·3 min read
Early work on monitorability evaluationsMETR published early work on evaluating monitorability of AI systems.
eval
SafetyMETR·245d ago·3 min read
Common Elements of Frontier AI Safety Policies (December 2025 Update)METR updated documentation on common elements of frontier AI safety policies as of December 2025.
ResearchMETR·265d ago·3 min read
GPT-5.1-Codex-Max Evaluation ResultsMETR published evaluation results for GPT-5.1-Codex-Max model.
evalgpt
ResearchMETR·287d ago·3 min read
Review of the Anthropic Summer 2025 Pilot Sabotage Risk ReportMETR reviews Anthropic's Summer 2025 pilot sabotage risk report.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.