#019 — Graphify: 70× Fewer Tokens or Marketing?
Date: September 18, 2026 | 19:00 London / 21:00 Kyiv
Format: Solo episode · controlled experiment: Claude Code vs Claude Code + Graphify · ~90–120 minutes
Guest: —
YouTube: https://youtube.com/live/Vn57Gw4Txz4
Participants
Oleh Levchenko — Senior Full Stack JavaScript Developer (React / Node.js / AWS)
The AA · Cardiff, UK
YouTube @zloyleva · itfriday.community
What we'll cover
Part 1 — Where the 71.5× claim comes from
- Graphify promises a knowledge graph for your codebase and a bold claim — 71.5× fewer tokens
- We look at how much hype has already built up around Graphify on YouTube
- We explain what the number actually means: graph query vs raw corpus on a specific 52-file workload, not an end-to-end coding-task benchmark
- The main question: what's left of that number in a real Claude Code workflow
Part 2 — Live controlled experiment
- A real open-source repository, Excalidraw, and a real issue, #11846 — "Why it is not possible to have 2 different files opened?"
- Two independent clones, same commit, same model, same Claude Code version, same prompt
- Two parallel terminals: Claude Code alone vs Claude Code + Graphify
- No hints, no transferring discoveries between runs — only what Claude finds on its own
- Graphify cold start measured separately — how much it actually costs to build the graph
Part 3 — The numbers and the verdict
- 🔥 TOKENS — the real savings, and how close they get to 70×
- ⏱ TIME — how long each run actually took
- ✅ SOLVED — whether the issue is genuinely fixed and validation passes
- The final question: is it worth keeping Graphify in a real Claude Code workflow?
Key takeaways
Added after the stream.
Timecodes
Added after the stream.