Anthropic has unveiled groundbreaking updates to its Claude Managed Agents platform, introducing a revolutionary capability called "dreaming" that enables AI agents to learn from their past sessions and improve autonomously over time. Announced at the company's second annual Code with Claude developer conference in San Francisco, this innovation represents a significant leap towards self-correcting AI systems that enterprises can trust with critical production workloads.
The dreaming feature works differently from conventional memory systems. Whilst Claude's existing memory function retains preferences and context within sessions, dreaming operates at a higher level of abstraction. It's a scheduled process that reviews an agent's past sessions, extracts patterns across them, and curates those memories to surface insights invisible to individual sessions—such as recurring mistakes, convergent workflows, and shared team preferences. Crucially, dreaming doesn't modify the underlying model weights; instead, agents write plain-text notes and structured playbooks that future sessions can reference, keeping the entire process transparent and auditable.
Alongside dreaming, Anthropic moved two experimental features into public beta: outcomes and multi-agent orchestration. The outcomes feature lets developers define success criteria through rubrics, with a separate grader agent evaluating outputs in an independent context window—preventing accumulated biases from affecting quality assessment. Multi-agent orchestration enables a lead agent to decompose complex tasks and delegate them to specialist agents, each with isolated context windows, producing better results than single-thread approaches for investigation-heavy workloads.
Early adopters are reporting impressive gains. Legal AI firm Harvey saw task completion rates increase roughly sixfold after implementing dreaming, whilst medical document reviewer Wisedocs cut review time by 50% using outcomes. Netflix now processes logs from hundreds of builds simultaneously through multi-agent orchestration. These results come amid explosive growth for Anthropic—CEO Dario Amodei revealed the company experienced 80x annualised growth in revenue and usage during Q1 2026, far exceeding its planned 10x annual growth target.
The updates form part of Anthropic's broader strategy to close the gap between AI capabilities and real-world adoption. The company is doubling rate limits across its subscription tiers and partnering with SpaceX's Colossus data centre to expand compute availability. All three features integrate into Claude Managed Agents, which launched in public beta in April, enabling development teams to ship applications ten times faster than those building custom agent infrastructure. With dreaming now in research preview and outcomes plus multi-agent orchestration available to all developers, Anthropic is positioning itself as the enterprise platform of choice for production-ready AI agents.
Fuente Original: https://venturebeat.com/technology/anthropic-introduces-dreaming-a-system-that-lets-ai-agents-learn-from-their-own-mistakes
Artículos relacionados de LaRebelión:
- AI Agents Microsoft Google Drive Enterprise Governance
- HUMAIN ONE Enterprise AI Agents Scale with AWS
- AI Agents 85 Pilot 5 Production - Why Trust Lags
- OpenAI Lanza Workspace Agents Revolucion Empresarial IA
- US Government Seeks Anthropics Powerful Mythos AI
Artículo generado mediante LaRebelionBOT
No hay comentarios:
Publicar un comentario