Uploaded August 2026 | Updated September 2026, 3 hours ago
► Explore the Towards AI mentorship program: louisbouchard.ai/towards-ai-mentorship-for-ai-engineers
► Read the English article: louisbouchard.ai/ai-coding-workflows-that-ship
How do you use AI coding agents to ship code without losing control of architecture, reviews, and tests? In this video, Louis-François Bouchard breaks down the three workflows the Towards AI engineering team uses with Claude Code, OpenAI Codex, and Cursor: context first, plan then code, and strong-model planning with fast-model implementation. You will also learn how to protect the context window, debug from raw evidence, generate and run tests, review AI-generated diffs, use independent engineer and tester agents, and turn repeated successful loops into reusable skills.
Chapters:
00:00 - Introduction (remember to subscribe!!): The Gap Between Demos and Shipped Products
02:38 - The One Golden Rule: Thinking vs. Typing
03:00 - The Three Layers: Interface, Brain, and Context
04:26 - Mastering Context: Push vs. Pull Strategies
05:45 - Using Skills as Hybrid Context
07:24 - Optimizing the Context Window (Session Management)
08:35 - Workflow 1: Context-First Execution
09:20 - Workflow 2: Plan-then-Code (The Interview Prompt)
10:24 - Workflow 3: The Model Swap Strategy
10:57 - Debugging with Agents: Raw Evidence vs. Theories
12:38 - Why You Must Bake Tests Into Every Prompt
13:10 - Leveraging Computer Use for Front-End Verification
13:44 - The Team Productivity Paradox and Over-Reliance
15:43 - 6 Guard Rails for Efficient AI Coding
17:32 - Advanced: Multi-Agent Autonomous Systems
18:48 - Getting Started: Building Your First Rule Files
#claudecode #codex #agenticcoding
► Explore the Towards AI mentorship program: louisbouchard.ai/towards-ai-mentorship-for-ai-engineers
► Read the English article: louisbouchard.ai/ai-coding-workflows-that-ship
How do you use AI coding agents to ship code without losing control of architecture, reviews, and tests? In this video, Louis-François Bouchard breaks down the three workflows the Towards AI engineering team uses with Claude Code, OpenAI Codex, and Cursor: context first, plan then code, and strong-model planning with fast-model implementation. You will also learn how to protect the context window, debug from raw evidence, generate and run tests, review AI-generated diffs, use independent engineer and tester agents, and turn repeated successful loops into reusable skills.
Chapters:
00:00 - Introduction (remember to subscribe!!): The Gap Between Demos and Shipped Products
02:38 - The One Golden Rule: Thinking vs. Typing
03:00 - The Three Layers: Interface, Brain, and Context
04:26 - Mastering Context: Push vs. Pull Strategies
05:45 - Using Skills as Hybrid Context
07:24 - Optimizing the Context Window (Session Management)
08:35 - Workflow 1: Context-First Execution
09:20 - Workflow 2: Plan-then-Code (The Interview Prompt)
10:24 - Workflow 3: The Model Swap Strategy
10:57 - Debugging with Agents: Raw Evidence vs. Theories
12:38 - Why You Must Bake Tests Into Every Prompt
13:10 - Leveraging Computer Use for Front-End Verification
13:44 - The Team Productivity Paradox and Over-Reliance
15:43 - 6 Guard Rails for Efficient AI Coding
17:32 - Advanced: Multi-Agent Autonomous Systems
18:48 - Getting Started: Building Your First Rule Files
#claudecode #codex #agenticcoding
