Materials for a 4-hour workshop and a 30-minute conference keynote on the move from vibe coding to agentic engineering.
By Michael Hunger (@jexp, VP Product Innovation / AI at Neo4j).
.
├── workshop/ # 4-hour workshop, ~16 section docs + outline
│ ├── workshop-outline.md ← start here for the full running order
│ ├── workshop-intro.md
│ ├── workshop-exercise-1.md
│ ├── workshop-harnesses.md
│ ├── workshop-tool-use.md
│ ├── workshop-cost-tokens.md
│ ├── workshop-benchmarks.md
│ ├── workshop-practices.md
│ ├── workshop-own-experience.md
│ ├── workshop-mental-safety.md
│ ├── workshop-exercise-2.md
│ ├── workshop-resources.md
│ └── agentic-engineering-ideas.md (raw notes)
│
├── keynote/ # 30-min keynote drafts
│ ├── prompt.md (original brief)
│ ├── vibecode-keynote-v1.md
│ ├── vibecode-keynote-v2.md
│ └── vibecode-keynote-v3.md
│
├── research/ # Source / reference material that fed the workshop
│ ├── coding-agents/ ← per-tool deep dives + comparison tables
│ ├── context-engineering-mining.md ← reusable bits from prior CE deck
│ ├── context-engineering-presentation.pdf ← prior CE deck (40 slides)
│ └── agent-hooks.md ← hooks comparison across harnesses
│
├── assets/ # Images, QR codes, etc.
│ └── qrcode-github-jexp-agentic-engineering.png
│
├── CLAUDE.md # Repo memory file for Claude Code / agentic editors
├── progress.md # Running log of what's been written and what's next
└── README.md # This file
- Reading the workshop? Open
workshop/workshop-outline.md— it's the master running order, points at every other workshop file, and includes timing + slide budgets. - Reading the keynote? Open
keynote/vibecode-keynote-v3.md— current draft. - Comparing agentic coding tools? Open
research/coding-agents/README.md— 16 per-tool deep dives + comparison tables for Claude Code, Codex, Cursor, Aider, Cline, Copilot agent mode, Junie, opencode, Goose, Crush, Pi, Gemini CLI, Antigravity, Windsurf, Devin, Kiro.
The short version (the long version is the keynote):
- Vibe coding (Karpathy, Feb 2025) is real and useful — for prototypes, learning, throwaway tools.
- Agentic engineering is the next phase for production work — and it brings the strong engineering disciplines back, not as overhead, but as the way agents stay useful.
- Good software design was never for the machine. It was always for other developers who had to read, maintain, and extend the code. Now those developers include AI teammates that learned the craft from human-written code.
- The system you build around the model is the moat. The model gets better every six months. The system is yours.
Materials in this repo are released under CC BY 4.0. Use, adapt, share — please credit.
Code snippets within docs are released under MIT.
- Email: michael@neo4j.com
- GitHub: @jexp
- Talk to me about: graphs + LLMs, agent memory, MCP, context engineering