Graph Engineering: replace giant prompts with parallel agents wired as a graph
Pavan_Belagatti · x · 2026-11-15
The author argues most agentic AI workflows share a design flaw: developers write one giant prompt and hope the model gets everything right in one go — a bad pattern for agentic systems. His alternative is Graph Engineering:
- Design the AI system as a graph: every step of work is a node, every path between steps is an edge
- Agents run in parallel, and feedback goes only to the parts that need it, not back through the whole loop
- He built an incident-investigation demo where three agents check metrics, logs, and deployment history simultaneously, then a correlation agent finds the likely cause and recommends a fix
- A hands-on guide and video walkthrough are available
Related event: Graph Engineering: Replacing Giant Prompts with Multi-Agent Workflows(2 posts)→
More from coding & agent
- Burkov Asks: Any Open-Weight Model + Harness Matching Codex or Claude Code? — burkov · 2026-11-15
- Coding over image models: pure-code art generation in Gemini and the app's nano banana bias — andrew_n_carr · 2026-11-15
- Tailscale + Claude Code trick: preview links that open on your phone — viksit · 2026-11-15
- Open-source MCP server integrates OpenAI API with dynamic plugin tools — modelcontextprotocol · 2026-11-15
- Openfda MCP connector exposes FDA drug safety data to AI agents — modelcontextprotocol · 2026-11-15
- Multiplayer AI's potential extends well beyond software development — diegobuildsai · 2026-11-15