Microsoft unveiled Scout, an always-on AI agent embedded in Teams and Microsoft 365 that reads messages, drafts responses, and chases tasks without waiting for a user prompt.

• Persistent ambient operation distinguishes Scout from reactive copilots: it monitors workspace context continuously and acts while the user is offline, collapsing the gap between notification and action.
• Announced by CVP Omar Shahine, Scout sits inside the existing M365 subscription tier, positioning Microsoft to defend seat count against point-solution agents from OpenAI and Anthropic.

Visual Studio Code 1.123 ships today with cross-device chat session sync, an enlarged context window for select models, and parallel agent window support.

• Cross-device sync ties agent chat history to a user's GitHub account by default, meaning developers can resume mid-task agent sessions on any machine without context loss.
• Parallel agent windows let engineers run concurrent coding agents on separate branches or tasks simultaneously, a workflow that until now required separate IDE instances.

Google added native Gemma 4 Multi-Token Prediction support to LiteRT-LM, its on-device LLM runtime, delivering up to 2.2x faster decoding and 3.7x faster prefill versus competing local inference frameworks.

• MTP decoding gains reach 1.6x for Gemma 4 E2B and 2.2x for Gemma 4 E4B, with prefill improvements of 1.8x to 3.7x compared to llama.cpp, MLX, Cactus, and ONNX Runtime benchmarks cited by Google.
• On-device target is a 16GB laptop, making Gemma 4 12B with audio, image, and video processing viable without cloud dependencies, a direct challenge to Nvidia's RTX Spark local inference pitch.

Radiant Logic released new agentic AI capabilities for its Identity Data Fabric platform, adding continuous risk scoring and cross-platform agent visibility, timed to Identiverse 2026.

• Platform-agnostic coverage means Radiant Logic claims to surface identity risk across any agent framework, positioning against Okta and Microsoft Entra as agentic workloads multiply non-human identities.
• Continuous risk scoring replaces periodic access reviews with real-time posture signals, directly addressing the IAM gap that Verizon's 2026 DBIR flagged as vulnerability exploitation overtaking stolen credentials as the leading attack vector.

Anthropic disclosed that as of May 2026, Claude authors more than 80% of its own production code, with engineers shipping 8x more code per quarter than they did in 2024.

• One Anthropic engineer has not written a line of code in five months, with Claude handling the full implementation workload, a data point that reframes productivity benchmarks organizations are using to justify AI coding tool spend.
• The Anthropic Institute paper accompanying the disclosure maps a credible path to recursive self-improvement and calls for a verifiable global pause mechanism if systems begin improving faster than governance can track, adding regulatory risk context to the productivity story.

CrowdStrike reported Q1 2026 net new ARR of $256M, up 32% year over year, alongside $468M in free cash flow, with management attributing the acceleration to AI-driven detection demand.

• Full-year guidance raised following Q1 results, with CEO positioning AI-speed attacks as the core demand driver: the company now frames its platform as the only response capable of matching AI-generated threat velocity.
• Stock dropped post-earnings despite the beat, with analysts citing valuation multiple compression as AI security becomes table stakes rather than premium differentiation, a dynamic that mirrors Okta's post-Q1 trajectory.