Podcast Summary

AI and I with Dan Shipper

Episode: How Two Engineers Ship Like a Team of 15 With AI Agents

Guests: Kieran Klaassen & Nityesh Agarwal

Date: June 11, 2025

Episode Overview

In this episode, Dan Shipper dives deep into disruptive engineering workflows with Kieran Klaassen (GM) and Nityesh Agarwal (Engineer), both of whom are building Quora—the AI email assistant from Every. Together, they explore how two engineers leverage advanced AI agents to produce output equivalent to a team of 15, fundamentally changing how software is built. The episode details their step-by-step workflows, experiments with various coding agents (especially Claude Code by Anthropic), prompt engineering, parallelization, and the evolving role of humans as managers of AI coding "workers."

Key Discussion Points and Insights

1. What Is “Compounding Engineering”?

• Compounding engineering: With every task completed using AI agents, the process itself becomes easier and more automated the next time. (00:00–00:49, 23:27)
• Key quote:

  “With each piece of work you do, you’re making it easier to do the next piece of work.” — Dan Shipper (01:07)

2. AI Is Doing More Than Just Coding

• AI can (and should) be used for everything: Not just for writing code, but also for research, workflow automation, and project management.
  (00:06, 04:38–06:12)
• Quote:

  "Coding with AI is more than just the coding part. Utilizing it for research, for workflows. It should be used for everything." — Kieran Klaassen (00:06, 04:38–06:12)

3. The Claude Code Workflow

• Claude Code explained: CLI tool from Anthropic, allowing not just code generation but file navigation, web search, GitHub operations, and more—all agentically.
  (06:35–09:52, 10:50–11:52)
• Engineers operate as AI managers—delegating tasks, reviewing outputs, and orchestrating parallel “workers.”
• Claude Code is praised for simplicity and flexibility, working well even for non-technical users.

4. Practical Demo — Running Tasks in Parallel

• Kieran demonstrates delegating multiple tasks (e.g., checking shipped features, generating marketing copy, planning new features) simultaneously to AI agents.
  (09:52–14:49, 15:33–19:28)
• Memorable moment:

  “We had, I think, six or seven [agents] running at the same time because we were just like, new idea, let’s go. New idea, let’s go.” — Kieran Klaassen (28:25)

5. Prompt Engineering as a Force Multiplier

• The team created “prompts that build prompts”—meta-prompts that generate detailed GitHub issues from a single voiced feature idea.
  (19:28–25:06)
• Used voice-to-text and even unreleased internal tools (Monologue) for low-friction feature specification.
• Leverage Anthropic’s Prompt Improver for better research prompts and documentation (23:27–25:06).
• Key insight:

  "What you did first is spent time building a prompt that effectively builds other prompts." — Dan Shipper (25:06)

6. The Human-in-the-Loop: Skill, Taste, and Intuition

• Human review remains critical: catching AI’s errors as early as possible is essential for leveraging AI at a high level.
  (28:25–30:18, 34:22–36:49)
• Quote:

  "You want to catch those problems early—check the AI’s work at the lowest value stage." — Nityesh Agarwal (52:44)

7. Parallelized, Social Coding and “Manager of Agents” Model

• Coding is now more parallel, social, and collaborative—features ship “live” during conversation.
• Engineers increasingly operate as orchestrators of agent pipelines rather than writing code directly.
  (25:06–28:25, 30:18–32:33)
• Quote:

  “It’s a kind of more social way to code. Like, we’re coding right now, building stuff, which was not possible before.” — Dan Shipper (25:06)

8. Problems and New Engineering Practices in Agentic Workflows

• Strong need for early validation and planning. If the “plan” is wrong, agentic code generation can go far in the wrong direction.
• Use of unit tests, smoke tests, “evals” for prompts, and iterative correctness checks becomes even more essential.
• Quote:

  “AI ... can now do so many things for us, it has become really important to focus on the earliest part of things.” — Nityesh Agarwal (34:22)

9. Comparing and Ranking Coding Agents (Tier List)

• Claude Code: S-tier (“takes the cake”)
• Amp: S-tier (“just gets work done”)
• Friday: S/A-tier (especially for UI work)
• Cursor: A-tier with Claude 4
• Charlie: B/A (excellent code reviews)
• Devon, Codex, Factory: Bs and Cs (context-specific)
• Copilot: D (lacks agentic capabilities—“not my thing”)
• Memorable moment:

  “Cursor is very good with CL4.” — Kieran Klaassen (42:56)
  “Amp ... S-tier under Claude Code ... very good at just getting work done.” — Kieran Klaassen (47:42)

10. Tools, Tips, and Takeaways

• Try Claude Code—even if you’re not a coder. The guests observe non-engineers successfully adopting it if they’re willing to try. (52:00)
• Always review AI’s work at the “lowest value stage” to catch errors before they escalate.
• The best results require both experimentation and human taste/intuition alongside rapid agentic workflows.
  (33:56, 36:49–38:41)
• Continuous learning: the landscape changes every 3 months.

  “…the entire coding landscape just changes completely every three months and you realize nobody’s at the forefront.” — Nityesh Agarwal (32:33)

Notable Quotes & Memorable Moments

| Timestamp | Speaker | Quote | |-------------|-------------|------------------------------------------------------------------------------------------------------------| | 00:06 | Kieran | "Coding with AI is more than just the coding part. Utilizing it for research, for workflows. It should be used for everything." | | 04:38 | Kieran | "...the biggest thing is a realization in myself that coding with AI is more than just the coding part...it should be used for everything." | | 10:08 | Kieran | "...we've been really leaning into, let's AI do the work for us, and we're just managing the AI." | | 14:49 | Dan Shipper | "...that task could take anywhere from like 30 minutes to a couple hours without, without AI...and now you just sort of like send off requests like that and then you can send off another one and another one..." | | 19:28 | Dan Shipper | "Kieran almost never types anything and does all voice to text...it's just a different way to code." | | 23:27 | Kieran | "This is part of the compounding effect. It's like having an idea that has a lot of outcomes." | | 25:06 | Dan Shipper | "What you did first is spent time building a prompt that effectively builds other prompts." | | 28:25 | Kieran | "We had, I think, six or seven [agents] running at the same time because we were just like, new idea, let's go. New idea, let's go." | | 30:05 | Kieran | "There's no magic prompt that does everything. Like it is about using it the right way and using it to its strengths." | | 32:33 | Nityesh | "...the entire coding landscape just changes completely every three months and you realize nobody's at the forefront." | | 36:49 | Kieran | "It's kind of boring to read most of the time, but you can make it more fun.... [but] then it misses things again. So it's actually important." | | 41:42 | Dan Shipper | "I want to spend five minutes with Kieran doing a S-tier through F-tier ranking of agents." | | 47:42 | Kieran | "Amp ... S-tier under Claude Code ... very good at just getting work done." | | 52:44 | Nityesh | "Just be sure to check the AI's work at the lowest value stage. You want to catch those problems early." |

Timestamps for Key Segments

• Compounding Engineering Explained — 00:00–00:49, 23:27
• AI Beyond Coding — 04:38–06:12
• Claude Code Demo & Capabilities — 06:35–11:52
• Running Parallel Agents & Managing Output — 09:52–14:49, 15:33–19:28
• Prompt Engineering Meta-Workflow — 19:28–25:06
• Human-in-the-Loop, Review & Taste — 28:25–30:18, 34:22–36:49
• Problems & New Engineering Practices — 34:22–38:41
• Tier List: Coding Agents Ranking — 41:42–49:14
• Final Advice, Takeaways — 52:00–53:09

Conclusion & Takeaways

This episode demystifies what it looks like to operate "at the bleeding edge" of AI-driven software engineering. Kieran and Nityesh demonstrate a future already arriving—where most engineering work becomes orchestrating multiple AI “workers,” designing workflows and meta-prompts, and applying human discernment at key checkpoints. For both seasoned and new programmers, the episode is a guide to integrating agentic tools, designing compounding processes, and embracing the rapid, near-chaotic evolution of the AI engineering landscape.

Final advice from the guests:

• Try Claude Code and don’t be intimidated by technical tools.
• Always check AI’s outputs early and rigorously.
• Embrace a manager-of-agents mindset and push yourself to experiment—because “the entire coding landscape just changes completely every three months.”