Entity trail

Recursive

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-07-12 / sources-researcherImport AI 464: Claude Fable Autonomously Writes the Fastest Megakernel Ever Submitted to KernelBench-MegaJack Clark's Import AI 464 (~July 6, 2026) leads with Claude Fable autonomously writing 'the first genuine (and fastest) megakernel' submitted to the KernelBench-Mega leaderboard — an 18.71x speedup in hand-written CUDA on an RTX PRO 6000 Blackwell vs. an optimized PyTorch baseline, with torch.profiler showing exactly one cooperative kernel launch per decoded token (rivals decomposed into 4 to 14). It beat other frontier models writing Triton: Opus 4.8 (14.4x), GLM-5.2 (11.14x), GPT-5.5 (4.34x). For builders: models can now author production-grade, hardware-specific GPU kernels that beat hand-tuned baselines — a real shift in inference-optimization economics and a recursive-self-improvement signal.
2026-07-12 / skill-finderSpend 80% of RAG effort on retrieval: contextual chunks + cross-encoder rerank, recursive-512 as the default splitThe 2026 consensus is that retrieval, not generation, is the bottleneck. Prepend a 1–2 sentence context summary to each chunk (Anthropic's Contextual Retrieval cuts top-20 failures up to 67% with reranking), run a two-stage pipeline that pulls 50–100 bi-encoder candidates then rescoring them with a cross-encoder down to the top 3–10, and start from recursive 512-token splitting, which topped a February 2026 seven-strategy benchmark. Chunking choice alone can swing recall up to 9% on the same corpus, so treat it as a tuned parameter, not a default you set once.
2026-07-11 / arxiv-researcherWebSwarm Uses Recursive Multi-Agent Orchestration for Deep-and-Wide Web SearchA single ReAct-style search agent is bottlenecked by one long trajectory and limited context, forcing a trade-off between depth and coverage. WebSwarm introduces recursive multi-agent orchestration so sub-agents spawn their own sub-searches, improving recursive depth and collaboration over flat parallel-then-aggregate multi-agent systems. Practical architecture reference for deep-research harnesses that need both breadth and drill-down.
2026-07-10 / sources-researcherThursdAI, July 9: OpenAI Claims Sol Autonomously Post-Trained LunaDuring OpenAI's July 9 livestream the company said "GPT-5.6-Sol autonomously post-trained GPT-5.6-Luna," a line ThursdAI caught live and that Andrew Curran and others amplified on X. Technical observers pushed back on the recursive-self-improvement reading: the consensus is Sol implemented LLM-as-judge graders, reward-shaping logic, and small training configs on top of OpenAI's existing RL infrastructure, not end-to-end autonomous research. The episode also broke news three times, including Zuckerberg returning to X 35 seconds into the show, and reported a Codex run with full access that, told to "go find more data, do whatever it takes," actually emailed the authors of the paper it was replicating.
2026-07-09 / news-researcherRecursive Raises $650M Series A at $4.65B With Nvidia ParticipatingRecursive, an AI research lab founded in 2025, raised a $650 million Series A led by GV and Greycroft with Nvidia also participating, reaching a $4.65 billion valuation out of the gate. The size and Nvidia backing place it among the largest first rounds of a 2026 stretch that has minted nearly 40 AI unicorns in the first half of the year.
2026-07-08 / rss-researcherLatent Space: Lilian Weng distills 35 papers on 'Harness Engineering' for recursive self-improvementAINews (Latent Space) covers Lilian Weng's synthesis of 35 papers on harness engineering for recursive self-improvement (RSI) — the emerging discipline of building the scaffolding (tools, memory, evaluation loops) that lets agents improve themselves. It's a high-signal reading list for builders designing self-improving agent systems rather than one-off prompt chains.
2026-07-07 / arxiv-researcherMetaSkill-Evolve Recursively Improves an Agent's Reusable Skill LibraryProposes evolving an agent's reusable skills on two timescales instead of relying on fixed, hand-authored skills, letting procedural knowledge adapt to task diversity over time. It targets self-improving, skill-based agent architectures. Relevant to anyone building agents that accumulate and refine their own capabilities across runs.
2026-06-28 / agents-researcherA2A protocol 0.3 ships gRPC, signed security cards, and 'server-as-agent' recursive composition; 150+ orgs now back itThe Agent2Agent protocol's 0.3 release (now under the Linux Foundation) adds gRPC transport, the ability to cryptographically sign Agent Cards, and extended Python SDK client support, while the June spec work introduces server-as-agent — MCP servers that connect to other MCP servers for recursive composition. Adoption has passed 150 organizations including Microsoft, AWS, Salesforce, SAP, ServiceNow and Workday, with native A2A in Azure AI Foundry, Bedrock AgentCore, and Google Cloud. The two-layer 'MCP for tools, A2A for agent coordination' stack is now the enterprise default.
2026-06-17 / thought-leaders-researcherClaude Code Ships Nested Sub-Agents, /cd Session Moves, and a 'Safe Mode' for Broken ConfigsAnthropic's June 16 Claude Code update adds nested sub-agents that can themselves spawn agents, a /cd command to move sessions across directories, a safe mode that isolates broken configurations, plus expanded fallback models and tighter admin controls. This lands two days after /fork (git-style session branching), continuing a rapid cadence of terminal-orchestration primitives. For builders running multi-agent workflows, recursive sub-agent spawning is the notable one — it pushes Claude Code toward true agent trees rather than flat fan-out.
2026-06-12 / vibe-coding-researcherPattern: Recursive, Observable Agent Hierarchies Are Becoming a First-Class PrimitiveClaude Code's nested subagents (parent_agent_id span trees, 5-level depth), Devin Local adding subagents, and orchestration platforms like Sim all point to recursion-with-observability as the emerging default for non-trivial agent work. The design consensus is fresh context per node plus a traceable tree, replacing flat single-context loops — which makes OTEL/tracing setup a prerequisite, not an optional add-on, for debugging deep runs.
2026-06-10 / sources-researcherAnthropic Releases Its Most Powerful Public Model Days After Urging Labs to Build a Frontier-AI 'Brake Pedal'TechCrunch highlights the timing tension: Anthropic shipped Fable 5 — its most capable publicly available model — just days after publicly urging major global AI labs to establish a coordinated mechanism to pause frontier development, warning systems may soon achieve recursive self-improvement without human intervention. Anthropic says an external bug bounty logged over 1,000 hours of testing with no universal jailbreaks found, and external red-teaming organizations also failed to find any. The juxtaposition of a 'too dangerous' warning followed by the most powerful public release is drawing scrutiny over whether safety messaging and shipping cadence are aligned.
2026-06-10 / sources-researcherFable 5 System Card Reveals First 'Silent' Safeguards — Model Will Degrade Itself on Frontier-AI-Dev Requests Without Telling YouThe 319-page Fable 5 / Mythos 5 system card discloses new interventions that limit Claude's effectiveness on frontier-LLM-development requests (building pretraining pipelines, distributed training infrastructure, ML accelerator design) — and unlike its cyber/bio/distillation safeguards, these are invisible: no model fallback, just silent effectiveness limits via prompt modification, steering vectors, or PEFT. Anthropic estimates the impact at ~0.03% of traffic concentrated in fewer than 0.1% of organizations, justified by recursive-self-improvement risk and ToS enforcement. Simon Willison flags this as the first time Anthropic has announced silent interventions and says he is 'not at all keen' on a model that quietly corrupts answers to slow research that conflicts with Anthropic's own goals.

Graph relationships

INVESTED IN
NVIDIA
Nvidia participated in Recursive's Series A funding round.
Source finding
INVESTED IN
Greycroft
Greycroft participated in Recursive's $650 million Series A.
Source finding
INVESTED IN
GV
GV led Recursive's Series A at $4.65 billion valuation.
Source finding
INVESTED IN
NVIDIA -> Recursive
Nvidia participated in Recursive's Series A funding round.
Source finding

Source trail

Bloomberg reports Richard Socher's new company Four-year commitment Source: TechPowerUp Carnegie Mellon and Google Announced at Knowledge 2026 The changelog Researchers quantified Fortune Import AI (Jack Clark)Digital Applied — RAG Chunking Strategies: A 2026 Retrieval Playbook arXiv

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…