| Episode | Status |
|---|---|
If you’re using AI to just write code, you’re missing out.Two engineers at Every shipped six features, five bug fixes, and three infrastructure updates in one week—and they did it by designing workflo...
Two engineers at Every demonstrate how they ship like a team of 15 using Claude Code and AI agents. They've developed a compounding engineering workflow where each task makes the next easier—using AI for research, planning, and implementation while maintaining human oversight at critical decision points. The episode includes a detailed walkthrough of their Claude Code workflow, custom prompts that generate detailed implementation plans, and a tier ranking of all major AI coding assistants.
Kieran and Nityesh explain how they've transformed Cora's engineering workflow to feel like a 15-person team with just two engineers. They introduce the concept of 'compounding engineering' where each piece of work makes the next piece easier through systematic use of AI agents for research, workflows, and implementation.
Detailed demonstration of Claude Code, Anthropic's terminal-based coding agent. Unlike traditional IDEs, it runs in the terminal with access to the entire computer, can execute commands, search the web, take screenshots, and integrate with GitHub. The team shows how they use it to query git history, generate product updates, and check project pipelines.
The team reveals their workflow for creating detailed GitHub issues using custom Claude Code commands. They built a prompt that generates other prompts—taking voice-to-text feature descriptions and automatically researching best practices, analyzing the codebase, and creating comprehensive implementation plans with minimal human input.
Before Claude Opus 4 launched, the team spent two hours creating 20+ issues to prepare for the superior model. This strategic preparation allowed them to immediately leverage the new model's capabilities for maximum productivity, demonstrating forward-thinking workflow design.
Live demonstration of debugging a production issue where a form wasn't being sent. Claude Code analyzed git history, identified the problematic code removal from 14 days ago, created a fix, generated a PR, and wrote a migration script—all with minimal human intervention and 'zero energy cost.'
Description of the actual day-to-day workflow: running multiple Claude Code instances in parallel, using voice-to-text for all inputs, coding socially while on calls, and managing AI agents rather than writing code directly. The team hasn't touched Cursor or Windsurf in three weeks.
The team competes to see who can keep Claude Code running longest (Kieran's record: 25 minutes, Nityesh: 8 minutes). This demonstrates Opus 4's unprecedented autonomy—running complex, multi-step plans without intervention, a qualitative leap from previous agentic tools.
Nityesh shares the most important realization from 'High Output Management': catch errors at the earliest, lowest-cost stage. With AI's power to execute quickly, it's crucial to validate direction during planning before implementation, as mistakes compound rapidly with autonomous agents.
The team discusses how to make AI-generated planning documents less boring and more useful. Instead of traditional PRD format, they prompt for user stories, questions a good PM would ask, and concrete examples—making review more engaging while maintaining thoroughness.
Despite AI's capabilities, traditional software practices remain essential. The team emphasizes smoke tests, automated testing, and evals (tests for prompts). Kieran demonstrates having Claude Code run evals 10 times, identify failures, and iteratively improve prompts until they pass consistently.
Kieran ranks every major AI coding assistant based on extensive testing. Claude Code and AMP take S-tier, Cursor ranks A-tier, while Windsurf drops to C-tier for lacking Claude 4. The ranking reveals that model quality now matters more than IDE features, and different agents excel at different tasks.
Case study of bringing in an infrastructure expert for a 2-hour consultation. The team recorded the conversation, fed it to Claude to generate implementation issues, had the expert review, then used Claude Code to implement—compressing two weeks of work into hours while leveraging specialized expertise.
Best of the Pod: Claude Code - How Two Engineers Ship Like a Team of 15
Ask me anything about this podcast episode...
Try asking: