Syed Fahad

Stop grepping: giving coding agents a knowledge-graph memory

Submitted Oct 28, 2026

Preferred format: 15 minutes (flash talk with slides) · Track: Applied AI

Coding agents like Claude Code and Codex still find their context the slow way: grep, open a file, read, repeat. On a real codebase that burns tokens, misses how files connect across modules, and produces confident answers nobody can trace back to a source. When we shipped Graphify, an open-source knowledge-graph memory layer for AI coding assistants, we started from a simple bet: give the agent a map of the repo instead of a flashlight.

This talk walks through what that took in practice. Graphify parses a repo locally with tree-sitter, builds a graph of code, docs and schemas, clusters it with Leiden, and exposes it to agents through a CLI, skills, PreToolUse hooks and an MCP server. I’ll cover what our first version got wrong, where the abstractions leaked across 20+ different assistants, and why every edge now carries a confidence tag (extracted, inferred or ambiguous) and a file:line citation. I’ll show a before-and-after with slides: the same question asked of Claude Code with plain grep, then with Graphify’s graph in the loop.

Key takeaways

  • Why flat text retrieval breaks down for coding agents, and when a graph is (and isn’t) worth it.
  • Practical patterns for turning an agent harness into shared team infrastructure: skills, hooks and MCP, plus the failure modes we hit.

Who it’s for
Engineers and tech leads using or building coding agents, platform and devtools teams rolling AI assistants out across a team, and anyone working on context engineering, RAG or MCP.

Bio
Fahad heads Developer Relations at Graphify Labs (YC S26), leading developer awareness, advocacy programs, and growing community across AI, tech, and developer productivity. He maintains OSS, helps teams adopt better workflows, and shapes community driven engineering culture. An explorer and eternal optimist, he loves networking and sharing knowledge with builders.

Draft slides: https://docs.google.com/presentation/d/1g24VYquzld22xqKO9KAIkZo2ec6JiJhT1_-eYmS03VA/edit?usp=sharing

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hosted by

Jumpstart better data engineering and AI futures