Scored Intelligence Stream

Anthropic ships computer-use tool for autonomous multi-step agents

New tool lets Claude operate software UIs directly, enabling agents that navigate CRMs and support consoles without bespoke integrations.

Anthropic News

AWS Bedrock adds prompt caching and cross-region inference GA

Prompt caching cuts input token costs up to 90% for repeated context; cross-region inference improves throughput for spiky enterprise workloads.

AWS Bedrock What's New

OpenAI Realtime API adds lower-latency voice with function calling

Sub-300ms speech-to-speech with tool calls makes production-grade voice agents viable for contact-center scenarios.

OpenAI Blog

OWASP releases 2026 Top 10 for LLM applications

Updated list elevates prompt injection via tool use and data exfiltration through agent memory as the top enterprise risks.

OWASP Top 10 for LLMs

LangGraph Platform GA for durable, stateful agent orchestration

Managed runtime adds persistence, human-in-the-loop checkpoints, and observability for long-running agent workflows.

LangChain Blog

GitHub ships agentic Copilot Workspace to general availability

Task-to-PR agent plans, edits across files, and runs tests, compressing routine engineering work.

GitHub Changelog

Enterprise survey: agent pilots stall on data readiness, not models

70% of stalled enterprise AI pilots cite fragmented data and unclear ownership rather than model capability as the blocker.

Enterprise AI Briefings

New open-weights model matches frontier quality on summarization

A permissively licensed model reaches near-frontier scores on long-context summarization benchmarks at a fraction of hosted cost.

OpenAI Blog