How Two Engineers Ship Like a Team of 15 With AI Agents | Kieran Klaassen, Nityesh Agarwal

Overview

This episode explores how two engineers, Kieran Klaassen and Nityesh Agarwal, leverage AI agents to achieve extraordinary productivity, shipping features and fixes at a scale that feels like a team of 15. They discuss their innovative workflows using tools like Claude Code, their mental models for compounding engineering, and rank various AI coding agents based on their effectiveness.

Notable Quotes

- You're figuring out how to do compounding engineering—each piece of work makes the next easier. - Dan Shipper, on the transformative nature of AI workflows.

- We’ve been leaning into letting AI do the work for us, and we’re just managing the AI. - Kieran Klaassen, on redefining engineering roles.

- Claude Code simplifies agentic coding by a factor of 10—it’s just a text box, but it works because the model is so capable now. - Nityesh Agarwal, on why Claude Code stands out.

🚀 Rethinking Engineering with AI Agents

- Kieran and Nityesh describe how AI agents like Claude Code allow them to focus on managing workflows rather than writing code manually.

- They use AI for everything from research to implementation, creating a compounding effect where each task builds on the last.

- Tools like Claude Code enable them to generate detailed GitHub issues, research best practices, and even create pull requests autonomously.

🛠️ Claude Code: A Game-Changer in Agentic Coding

- Claude Code, a CLI tool by Anthropic, integrates deeply with local directories, GitHub, and other tools, making it more versatile than competitors like Cursor or Windsurf.

- Kieran demonstrates how he uses voice-to-text to describe features, which Claude Code then translates into detailed implementation plans.

- The tool excels at simplifying workflows, enabling engineers to focus on high-level decision-making while AI handles execution.

📈 The Mental Model of Compounding Engineering

- The team emphasizes the importance of building prompts that generate other prompts, creating a self-reinforcing system of productivity.

- By investing time upfront in crafting reusable workflows, they reduce friction for future tasks.

- Kieran likens this to hiring specialized developers for specific tasks, with different agents excelling in unique areas like UI design or code reviews.

🤖 Ranking AI Coding Agents

- Kieran ranks various AI agents based on their capabilities:

- S-tier: Claude Code, AMP (for its developer-focused design and seamless execution).

- A-tier: Cursor (with Claude 4 integration), Friday (for its opinionated and effective workflows).

- B-tier: Devin, Codex, Charlie (great for code reviews).

- C-tier: Windsurf (lacking Claude 4 integration).

- D-tier: GitHub Copilot (outdated for agentic workflows).

⚡ Challenges and Best Practices in AI-Driven Development

- The team stresses the importance of catching errors early in the process, following the principle of fixing problems at the lowest value stage.

- Human intuition remains critical for reviewing AI-generated plans and ensuring alignment with project goals.

- They advocate for using traditional testing methods and evals to validate both code and prompts, ensuring reliability at scale.