Deep dives into the AI open-source ecosystem — quality scores, trends, and analysis across 220,000+ repos. https://mcp.phasetransitions.ai/insights/ 2026-04-05T12:38:19Z Graham Rowe https://mcp.phasetransitions.ai/insights/agentic-ball-of-mud/ 2026-04-05T12:38:19Z 2026-04-05T12:38:19Z

Claude Code's leaked source reveals the architectural ceiling of agentic AI: markdown memory, prompt orchestration, zero crash recovery. Brilliant for interactive work, breaking as agents go autonomous. We mapped who's building durable state, observability, and crash recovery for what comes next.

Graham Rowe https://mcp.phasetransitions.ai/insights/agent-native-docs/ 2026-04-03T12:39:51Z 2026-04-03T12:39:51Z

Agent-native documentation: why agents choose tools based on docs, the llms.txt movement, and what the MCP ecosystem data reveals about doc quality and adoption.

Graham Rowe https://mcp.phasetransitions.ai/insights/agent-native-infrastructure/ 2026-04-03T12:39:44Z 2026-04-03T12:39:44Z

Agent-native infrastructure: the emerging category of services designed for AI agents as first-class entities. Identity, email, credentials, payments, and gateways — 30+ repos across 25 subcategories.

Graham Rowe https://mcp.phasetransitions.ai/insights/domain-expert-builders/ 2026-04-03T12:39:38Z 2026-04-03T12:39:38Z

Domain experts are arriving on GitHub via AI coding agents. 27% of new AI repos in 2026 sustain 300+ commits/month from single-owner accounts. Data-driven analysis of the domain expert builder phenomenon.

Graham Rowe https://mcp.phasetransitions.ai/insights/llm-eval-landscape/ 2026-04-03T09:53:26Z 2026-04-03T09:53:26Z

LLM evaluation landscape in 2026: 1,159 repos across 9 subcategories, no dominant winner. Decision guide covering RAG eval, agent benchmarks, code eval, and production monitoring — with quality scores updated daily.

Graham Rowe https://mcp.phasetransitions.ai/insights/agent-platform-architecture/ 2026-04-02T13:17:42Z 2026-04-02T13:17:42Z

Building an agent platform means solving four problems: sandboxing, governance, orchestration, and the agent itself. Only sandboxing is mature. Here's the full landscape with quality scores.

Graham Rowe https://mcp.phasetransitions.ai/insights/crewai-dependency-audit/ 2026-04-02T08:42:32Z 2026-04-02T08:42:32Z

Strategic dependency audit of crewAI's 47 dependencies. 6 of 7 scored deps are #1 in their category — world-class choices. Agent memory is the one category still in flux. Plus: what we can't yet score and why.

Graham Rowe https://mcp.phasetransitions.ai/insights/claude-code-ecosystem/ 2026-04-02T06:58:44Z 2026-04-02T06:58:44Z

A practitioner's guide to the Claude Code ecosystem in 2026. 2,400+ repos tracked daily across memory, skills, MCP servers, session managers, and platforms. What to install, what to skip, and where the signal is hiding.

Graham Rowe https://mcp.phasetransitions.ai/insights/agent-memory-landscape/ 2026-04-01T19:36:00Z 2026-04-01T19:36:00Z

Agent memory in 2026: 977 repos across 5 domains, scored daily on quality. A decision guide covering mem0, claude-mem, Cognee, Memvid, and the architecture choices that actually matter for persistent AI.

Graham Rowe https://mcp.phasetransitions.ai/insights/perception-browser-automation/ 2026-04-01T19:35:51Z 2026-04-01T19:35:51Z

A decision guide for browser automation in 2026. Which scraping, agentic browser, and stealth tools are actually maintained? Quality-scored daily across 340+ repos. Firecrawl, browser-use, Patchright, curl_cffi, and what foundation models change.

Graham Rowe https://mcp.phasetransitions.ai/insights/agent-skills-architecture/ 2026-04-01T10:26:36Z 2026-04-01T10:26:36Z

Agent skills architecture in 2026: how Composio, OpenClaw, Antigravity, and 2,700+ repos are building the composable capability layer for AI agents. Quality-scored daily. What works, what's noise, and where the trust gap is.

Graham Rowe https://mcp.phasetransitions.ai/insights/obsidian-pkm-agents/ 2026-04-01T09:58:51Z 2026-04-01T09:58:51Z

AI agents for Obsidian in 2026: which PKM tools, MCP servers, and agent plugins actually work? Quality-scored daily across 700+ repos. khoj, Smart Connections, mcpvault, and what's still experimental.

Graham Rowe https://mcp.phasetransitions.ai/insights/agent-governance-landscape/ 2026-04-01T09:26:49Z 2026-04-01T09:26:49Z

Agent governance in 2026: which sandboxing, guardrails, and security tools are production-ready? Quality-scored daily across 1,263 governance repos. E2B, OpenSandbox, Microsoft's toolkit, and what's still missing.

Graham Rowe https://mcp.phasetransitions.ai/insights/embeddings-shortcut/ 2026-04-01T06:44:43Z 2026-04-01T06:44:43Z

Embeddings are cheaper, simpler, and more powerful than keyword search, regex, or Elasticsearch for most use cases. pgvector in your existing Postgres handles 90% of needs. Here's the practical guide.

Graham Rowe https://mcp.phasetransitions.ai/insights/voice-app-stack/ 2026-04-01T04:19:52Z 2026-04-01T04:19:52Z

The complete voice AI stack for developers: STT (just use Whisper), TTS (5 real options), intent handling, orchestration, and evaluation. Opinionated recommendations at every layer.

Graham Rowe https://mcp.phasetransitions.ai/insights/voice-ai-landscape/ 2026-03-31T06:58:27Z 2026-03-31T06:58:27Z

A decision guide for voice AI in 2026. Which TTS, ASR, and voice agent libraries are actually maintained? Quality-scored daily across 9,679 repos. ESPnet, whisper.cpp, Edge TTS, and what's next.

Graham Rowe https://mcp.phasetransitions.ai/insights/openclaw-ecosystem/ 2026-03-30T11:52:32Z 2026-03-30T11:52:32Z

How an AI open-source ecosystem forms in real time: OpenClaw went from viral launch to 1,299 repos in 8 weeks. Timeline, adoption data, and the pattern behind ecosystem formation.

Graham Rowe