Optimizing Codebases for Agents

Conference: Coding Agents: AI Driven Dev Conference Date: March 3, 2026 Location: Computer History Museum, Mountain View, CA Speaker: Shrivu Shankar (VP AI, Abnormal Security)

Your agent isn't broken. Your codebase is.

What This Is

This repo contains the materials for the "Optimizing Codebases for Agents" workshop. It includes a demo app in two versions (before and after optimization), an AI-readiness scorecard, and agent race narrator notes.

Prerequisites

• A laptop with a development environment
• An AI coding tool installed and authenticated (Claude Code, Gemini CLI, Codex CLI, or Cursor)
• Python 3.10+
• A repository you want to audit (open source or personal). If you don't bring one, you can audit the A/ demo app.

Quick Start

git clone https://github.com/sshh12/coding-agents-workshop.git
cd coding-agents-workshop

Run the "Before" app (Version A)

cd A
pip install -r requirements.txt
python app.py
# Open http://localhost:8000

Run the "After" app (Version B)

cd B
pip install -r requirements.txt
python manage.py run
# Open http://localhost:8000

Run tests (Version B only)

cd B
pytest

Repo Structure

coding-agents-workshop/
  README.md           # You are here
  scorecard.md        # AI-Readiness Audit (use during the workshop)
  race.md             # Agent Race narrator notes
  slides.html         # Workshop slides (speaker notes embedded as HTML comments)
  A/                  # "Before" -- messy demo app (deliberate anti-patterns)
  B/                  # "After"  -- agent-optimized (same functionality)

Workshop Materials

• Slides -- Full slide deck (press N for speaker notes, arrow keys to navigate)
• scorecard.md -- Score your codebase 0-9 across three dimensions. Bring this up during the audit sprint.
• race.md -- Narrator notes for the live agent race demo. Two terminals, same prompt, different codebases.

Run the Scorecard on Your Repo

You can have Claude Code audit any repo against the scorecard automatically:

Use web fetch to read https://raw.githubusercontent.com/sshh12/coding-agents-workshop/refs/heads/main/scorecard.md and then spawn tasks in parallel to perform and report an audit of the current repo

Paste this as a prompt to Claude Code inside the repo you want to audit.

The Demo App

Both A/ and B/ implement the same ML Experiment Tracker with dashboard:

• Track experiments (name, description, status)
• Log runs with hyperparameters and metrics (accuracy, loss, latency)
• Compare runs side-by-side with charts
• Dashboard with status badges, metric charts, and activity feed

The difference is how the code is organized. Version A is a realistic mess. Version B is optimized for coding agents.

License

MIT