AgentPages, an open-source project by developer Ido Rozin, offers a practical implementation of fully autonomous research and publishing using GitHub's own emerging agentic workflow platform. Users fork the repository, define their research interests in a plain-text profile file, and hand off the rest to a GitHub Agentic Workflow (gh-aw) running on a 12-hour cron schedule. Each cycle, the agent reads its instructions, queries the web via Tavily's search API, updates a structured knowledge base, regenerates an Astro static site, commits the changes, opens a pull request, and auto-merges it — all without human involvement. The resulting site is served directly via GitHub Pages from the repository's docs/ folder.

The underlying runtime, GitHub Agentic Workflows (gh-aw), is a joint GitHub Next and Microsoft Research initiative that lets developers write agentic workflows in natural-language markdown, which are then compiled into GitHub Actions lock files. At version v0.58.3 with over 8,200 commits and 4,100 GitHub stars as of March 2026, gh-aw has moved well beyond prototype territory. It supports four AI engine backends — GitHub Copilot (default), Claude Code (Anthropic), Codex (OpenAI), and Gemini CLI (Google) — with users switching engines by editing a single field and recompiling. AgentPages defaults to claude-sonnet-4-6 but the engine swap requires only a one-line config change.

The project uses git itself as <a href="/news/2026-03-15-openclaw-superpowers-self-modifying-skill-library-for-persistent-openclaw-agents">agent memory</a>. The agent/ folder persists knowledge base files and action logs across runs, creating a stateful research loop that compounds over time. It also shows how current <a href="/news/2026-03-15-localagent-v0-5-0-local-first-rust-mcp-runtime">agent infrastructure</a> stacks together: Tavily handles search (with a free tier of 1,000 credits per month covering light personal use), Astro provides a deterministic static build target that an LLM can reliably update programmatically, and gh-aw's companion Agent Workflow Firewall (AWF) and MCP Gateway handle network isolation and tool access management. Live example sites include a self-documenting AgentPages project site and a real-time geopolitical conflict tracker, demonstrating the pipeline's range across both technical documentation and fast-moving news summarization.

GitHub has been pitching what it calls "Continuous AI" — autonomous agent runs as first-class software collaboration primitives running inside existing developer infrastructure. AgentPages is a working demonstration of that thesis. Because gh-aw runs on standard GitHub Actions, it requires no managed compute outside what developers already have. That's a direct structural challenge to platforms like Replit Agent or Devin that require teams to adopt new runtime surfaces. A Show HN submission that wires together search, coding agents, static site generation, and CI/CD publishing in one repo, with a Tavily free tier doing the heavy lifting on search costs, is a meaningful signal of how cheap the entry point has become.