Anthropic
1.124
102
ESPO: Error-Structured Prompt Optimization via Diagnose, Diversify, and Stabilize
Evolutionary prompt optimizers such as GEPA suffer from prompt bloat: each iteration appends rules and caveats, producing prompts up to 3longer yet no more accurate. We trace this to three deficiencies - incomplete error observation, limited search diversity, and unreliable selection - and propose ESPO (Error-Structured Prompt Optimization), which decomposes prompt optimization into three phases: Diagnose clusters all training errors into structural patterns in one round; Propose generates candidates via four complementary strategies with independent biases; Select applies bootstrap stability selection. On seven public NLP benchmarks - Tweet, MMLU, GSM8K, HotpotQA, ScoNe, HoVer, and PUPA - ESPO improves average accuracy by +3.76 pp over the state-of-the-art (74.67% vs 70.91% for GEPA), matching or exceeding GEPA on every dataset while producing prompts 47% shorter (1,004 vs 1,878 chars) and faster at inference. Cross-model experiments across four additional student models (Gemma 3 12B, Mistral 14B, Qwen3 32B, Claude Haiku 4.5) show ESPO yields the best average accuracy on every model tested, with the largest gap on Qwen3 GSM8K (15.00% 91.40%). A generalization bound (Appendix) grounds each phase in a corresponding term of the test-time gap, and the ablation confirms a key prediction: adding diversity without bootstrap selection actually hurts performance (-1.20%).
Environment Evolution for Terminal Agents
Scaling interactive and verifiable environments is critical for training terminal agents. As frontier models become more capable, environments synthesized from scratch become less challenging and thus provide limited learning signals. Recent co-evolution methods iteratively synthesize environments near the model's learnable frontier based on weaknesses exposed during rollouts. However, their dependence on on-policy rollouts limits generalization and the continuous provision of learning signals as the model becomes stronger. In this paper, we propose environment evolution, which incrementally increases environment difficulty off-policy and schedules the evolved environments generation by generation during training to provide continuous learning signals. We derive three evolution directions that influence environment difficulty from the multi-turn learning objective and then implement evolution along these directions through a loop-engineered multi-agent harness. Quantitative rollout experiments with Hy4 preview, Claude Opus 5, and GPT-5.6 Sol show that environment evolution consistently produces more difficult environments. We validate its effectiveness on Qwen3.6-27B and Qwen3.6-35B-A3B through simple long-horizon RL training, improving their performance by 14.4 and 18.0 percentage points on Terminal-Bench 2.1, respectively.
AI-driven development lifecycle using Amazon Bedrock AgentCore
Engineering teams adopting the AI-Driven Development Lifecycle (AI-DLC) often struggle to turn concepts into working code. This post walks through two reference implementations on Amazon Bedrock AgentCore, Kiro, and Claude Code: an SQL-to-ER-diagram generator and a multi-agent code security analyzer that put the AI-DLC construction phase into practice.
Migrate agentic workloads to Amazon Bedrock AgentCore
An agent that works in a notebook is not an agent in production. This post walks through migrating a LangGraph customer support agent to Amazon Bedrock AgentCore in two stages: onto Runtime, Gateway, and Memory, then to model-driven planning on Strands Agents, retiring operational burdens along the way.
ChatGPT, Grok, and Claude all went down at the same time
OpenAI's ChatGPT, xAI's Grok, and Anthropic's Claude are all experiencing issues. At around 11AM ET, ChatGPT started returning error messages for users trying to use…
Modernizing and scaling support operations with generative AI on AWS
Learn how to build a generative AI-based support operations platform on AWS that converts training videos into structured SOPs, applies Retrieval-Augmented Generation to guide ticket resolution, and uses machine learning to predict SLA risk and prioritize work.
How an AWS team detects dashboard content failures at scale using Amazon Bedrock
Business intelligence dashboards can fail silently, showing blank, stale, or wrong data even when every infrastructure monitor reports healthy. Learn how an AWS team built an automated, AI-powered content validation solution on Amazon Bedrock that scans hundreds of dashboards and alerts owners, cutting mean time to detection from days to under an hour.
From code to diagrams: Agentic architecture documentation with Amazon Bedrock AgentCore
Learn how a global interdealer broker built an automated architecture documentation pipeline on Amazon Bedrock AgentCore that analyzes .NET code bases, generates architecture diagrams, and maintains searchable documentation through Amazon Bedrock Knowledge Bases and AWS CodePipeline.
Trinity: Agentic AI-powered transition planning for students with disabilities
Learn how University Startups and its AWS partner g/d/n/a scaled Trinity, a conversational AI solution for students with disabilities, into a serverless multi-agent architecture on Amazon Bedrock that produces IDEA-aligned transition plans for school districts across the US.