Research
FLEET turns repeated LLM sampling into memory-guided search
FLEET reports better LiveCodeBench Pass@32 and roughly 3× sample-efficiency by adding memory to repeated LLM decoding, with important deployment limits.
New Features & Tech Innovations
-
Agents
Microsoft turns Copilot into a runtime for persistent workplace agents
Microsoft adds Copilot Home, Code, Autopilot and Managed Runtime, creating a governed execution layer for persistent workplace agents.
-
Agents
Microsoft Foundry makes scheduled and event-driven agents a managed service
Foundry Routines now provide managed scheduled and event-driven agent execution with identity selection, run history and continuation patterns.
-
Agents
Cursor pushes coding agents past the pull request with Rollouts and Security Review
Rollouts and Security Review extend Cursor agents from pull requests into deployment health and exploit-path security review.
-
Research
AIDE² rewrites its own research-agent harness, but hidden evaluations decide what survives
AIDE² repeatedly rewrote its own research-agent harness in an eight-day experiment. Held-out tests suggest transfer, while the evaluation gate remains the key control boundary.
-
Research
Claude found an unusual enzyme system in DNA data; the wet lab is what makes the result matter
Anthropic says Claude agents found a previously uncharacterized repeat-associated enzyme system. Wet-lab evidence is promising, but ART's biological function remains unknown.
-
Agents
DeepSeek Harness 0.1.7 turns agent configuration into a runtime plugin surface
DeepSeek Harness 0.1.7 makes agent capabilities more dynamically composable. Aipolix examines the benefits and the new operational risks.
-
New Models
GPT-6 Sol and Luna make model routing a first-order agent economics decision
GPT-6 Sol and Luna introduce a 20x price spread, large context and different cost boundaries. Aipolix examines agent routing economics.
-
New Models
Claude Opus 5.5 cuts agent costs, but safeguard routing changes how teams should evaluate it
Claude Opus 5.5 cuts token and cache costs while sensitive requests may use fallback models. Aipolix examines the impact on agent evaluation.
-
New Models
Grok 4.7 keeps token prices flat, but agent economics depend on the full task
Grok 4.7 keeps basic API token rates unchanged, but independent tests show higher output consumption and provider-specific long-context pricing.
