MIT finds AI supply-chain agents cut costs 67% but introduce decision-variance amplification (agent bullwhip effect) across coordination layers — unresolvable by sampling alone.
physics-intern multi-agent framework: Gemini 3.1 Pro goes 17.7% → 31.4% on CritPt benchmark, new SOTA. Specialized teams self-correct and compute intermediate results.
Anthropic ships Agent View for Claude Code: a unified terminal dashboard for managing parallel sessions with status indicators, inline replies, and /goal.
Johns Hopkins April 2026: frontier LLMs fail ~60% on multi-tier agent instructions. Gemini 3.1 Pro peaks at 42%; GPT 5.4 below 40%; Opus 4.6 at 33%.
11x.ai raised $74M on ~$3M real ARR with 70–80% first-year churn. Artisan's LinkedIn got rate-limited. Both expose why monolithic SDR agents fail at scale.
Cofounder 2 launches: multi-agent infra orchestrating engineering, sales, marketing, ops, design agents in one company-wide system. 'Infrastructure for the one person billion dollar company.'
JP Morgan's 'Ask David' multi-agent research system is live: supervisor + specialized subagents + LLM-judge + HITL — the emerging enterprise agent pattern.
Controlled studies show multi-agent systems amplify baseline errors up to 17.2x. Single-agent + tool augmentation now the empirically-grounded default.
Stanford, Google/MIT, AHE, LangChain, and Unblocked all published this week: harness quality outperforms agent count as the primary agentic performance lever.
New 'Recursive Multi-Agent Systems' paper (arXiv 2604.25917) by Jiaru Zou on HuggingFace proposes a framework for agents that recursively spawn sub-agents.
Research formalizes 'diversity collapse': multi-agent LLM systems homogenize outputs due to structural coupling — brainstorming setups must be explicitly engineered for heterogeneity.
Meta's first model from its Superintelligence Labs scores 50.2 on HLE With Tools using parallel multi-agent inference at test time.
Anthropic's Project Deal experiment ran live agent-to-agent negotiations among 69 staff: 186 deals, $4,000+ volume, Opus outperformed Haiku — but participants couldn't tell the difference.
Anthropic ran a live two-sided agent marketplace with 69 employees: 186 deals, $4,000+ volume — and model quality (Opus vs Haiku) was invisible to human participants throughout.
Sakana AI's Fugu beta hits SOTA on SWE-Pro, GPQA-D, and ALE-Bench with dynamic frontier model orchestration via an OpenAI-compatible API.
Curated AI insights — sent when there's something worth your inbox.