Podcast Summary: How I AI

Episode: From journalist to iOS developer: How LinkedIn’s editor builds with Claude Code | Daniel Roth
Host: Claire Vo
Guest: Daniel Roth
Date: March 16, 2026

Episode Overview

Theme:
This episode explores how Daniel Roth, long-time journalist and current LinkedIn editor, transformed himself into a solo iOS app builder by leveraging generative AI (specifically Claude Code by Anthropic). Daniel shares the practical, detailed workflow he uses—without any prior engineering experience—to build, iterate, and ship production-grade apps using “dueling agent” personas. The conversation focuses on demystifying AI-assisted app development, covering everything from project management and prompting, to product mindset and the realities of solo shipping apps today.

Key Discussion Points & Insights

1. Daniel’s Transition from Journalist to AI-Powered iOS Developer

• Daniel’s Background:
  • Previously a business writer and editor (Fortune, Wired, Forbes)
  • Has witnessed multiple sea changes in technology and media, with parallels between the rise of blogs/digital publishing and democratized software creation via AI
  • Daniel expresses excitement at how AI removes barriers for non-engineers to build software directly
  • Quote:

    "When generative AI started making its way into coding and it became clear that any of us could build anything, I had the exact same idea [as with blogs]." (03:29)

2. Motivations and Mindset for Building

• Personal Pain Drives Product:

  • Daniel’s flagship app, Commutely, answers his NYC subway frustrations.
  • He’s emblematic of the new trend of “personalized software”—starting with your own life, then scaling to others.
  • Quote:

    "It was that perfect product market fit because I was the entire product." (07:02)

• New Roles for Non-engineers:

  • Daniel compares his “vibe coder” role to being a “picky customer,” not a strict PM or architect.
  • Quote:

    "All I am is a really picky customer… walking through this house and telling the architect: no, I want this room blue." (00:36)

3. The “Dueling Claude Code Agents” Workflow

Workflow Breakdown:

• Constant Contextual Tracking:
  • Ideas and feature requests are tracked in a standing Claude chat, with a table ranking their value and effort.
• Documentation & Memory:
  • Every feature, plan, and conversation is logged as markdown files—essential for maintaining context over weekend bursts.
  • Tip:

    "Everything gets logged. That is one tip I try to give anyone that is in my shoes." (09:47)

The Persona-Based System:

• Two Main Agents:

  • Bob (The Builder):

    • Instruction: Plan first, build modularly, document everything, and always seek code review before proceeding.
    • Has explicit prompt/persona for lean building and clear doc habits.
    • Notable Detail:

      Bob is allowed to create sub-agents that he manages, but those sub-agents cannot create further agents. (25:26)

  • Ray (The Reviewer):

    • Persona: Senior software engineer, obsessed with integrity, security, and trust.
    • Reviews and sometimes rejects Bob’s plans and code.
    • Principle:

      "Ray cannot spawn. Ray is Ray… only he can say yes or no." (25:48)

• User’s Role (Daniel): Breaks ties between Builder and Reviewer, injecting “customer” priorities.
• Process:
  • Bob creates a plan → passes to Ray → Ray critiques/approves → plan returns to Bob for building and iteration.
  • Manual friction (copy/paste) retained for learning and oversight:
    • Quote:

      "Putting a little bit of friction in the process, where you're actually forced to copy and paste, put it over, read what Ray says… does help with learning." (15:39)

Organizational Lessons Mirror Agent System:

• Daniel draws from his 15 years at a tech company, modeling Ray on real principal engineers who are the ultimate gatekeepers.
• Humorous Reflection:
  • Reference to adding more agents (Amy the PM, Joe the AE), mimicking real org charts. (13:33)

4. Technical Stack & Tools

• Started with Cursor (AI IDE), then moved to Claude Code via web/desktop app for flexibility.
• Xcode remains the end-stage for compiling, simulating, and shipping apps.
• Daniel admits to using only a fraction of Xcode and relies on prompted instructions (acting almost like a “human agent” for the bots).
  • Quote:

    "I'm basically the bot. I'm like, Claude has told me what to do." (27:34)

5. Prompting Principles and AI Mindset

• Prompt Structure:

  • Daniel resists over-personification of his AI agents (avoiding "please," "thank you"), emphasizing clear, direct instructions.
  • Parenting Analogy:
    • “Managing AI is like managing a really smart but hungover intern.” (19:07)
    • Assume best intentions but always remind them and double check.
    • Quote:

      "I assume the AI has best intentions but has to be reminded about how we work. So I don't yell, I'm pretty clear and I try to give it a little bit of grace." (36:41)

• Routine Review/Workflows:

  • Daniel shares a daily Copilot workflow for inbox/task triage:
    • “What did I drop the ball on?” check at the end of the day—all missed tasks, unanswered messages, and follow-ups are surfaced across work systems.
    • Quote:

      “At some point you start managing so many different projects, you're like, I know I'm dropping the ball. Just tell me what I am dropping the ball on.” (34:24)

Memorable Quotes & Moments

• On democratized building:

  "In the past I would have had to go convince an engineer or a PM… now I could suddenly start building." (03:29)

• Agentic management as real org life:

  "I'm able to do this because I've now worked inside of a tech company for 15 years… a lot of this is based on people I've worked with." (26:10)

• On solo learning via AI:

  "I'm not doing this just to build. I'm doing this to learn… the problems I find I can solve uniquely as a human are things like, ‘should we prioritize this or not?’" (15:54)

• On UX and product voice:

  “As the picky customer, I could say, this is the voice of this thing. Make it funny, and wrote some copy for that.” (24:28)

Important Timestamps

• 00:36 — Daniel on rethinking his role as a builder/vibe coder
• 03:29 — Daniel’s origin story: from blogs to AI coding
• 07:02 — Building Commutely as his own core user/personal PMF
• 09:47 — Logging everything for continuity and accountability
• 11:35 — Bob and Ray agent personas; prompt strategies
• 14:20 — Manual hand-off and learn-by-doing rationale
• 19:07 — The "hungover intern" analogy for AI management
• 25:26 — Rules for spawning/nesting agents
• 27:34 — Admitting non-expert status, acting as “Claude’s bot”
• 34:24 — Copilot for “what did I drop the ball on?” workflow
• 36:41 — Prompting philosophy: assume best intentions, no personification

Practical Tips & Workflows to Copy

• Keep a Running Feature/Idea Tracker:
  • Use a persistent Claude chat for collecting, detailing, time/growth ranking, and storing ideas.
• Document Everything in Markdown:
  • Every interaction, plan, and feature goes in .md files—essential for part-time builders.
• Dueling Agent Principle:
  • Use two Claude personas: Builder (pushes ahead) and Reviewer (pushes back). Manually mediate between them for learning and QA.
• End-of-Day Copilot Review:
  • Ask your AI: “What did I drop the ball on?”—get actionable task lists to clear your plate before tomorrow.
• Be Direct in Prompts:
  • Ditch polite fluff; state roles, context, and memory recalls clearly for reliable, repeatable outputs.

Additional Resource Links

• Daniel’s LinkedIn: Daniel Roth
• Newsletter: Forward Deployed Editor—focused on AI for non-coders and lessons learned

Takeaways for Listeners

• Anyone—even with non-technical backgrounds—can build real apps using AI tools with the right workflow and learning orientation.
• Agentic thinking and AI prompting blend real-world management experience with modern product building (“managing your bots as a team”)
• Documenting, reviewing, and understanding AI output is as important as the build itself—don’t skip the learning!
• The biggest roadblocks are often not technical, but process (App Store, distribution, user feedback).

[For full episode and more resources, visit: howiaipod.com]