The Pragmatic Engineer Podcast

Episode: The Creator of Clawd: "I Ship Code I Don't Read"
Host: Gergely Orosz
Guest: Peter Steinberger (creator of Cloudbot; founder of PSPDFKit)
Date: January 28, 2026

Overview

In this episode, Gergely Orosz interviews Peter Steinberger, creator of the emerging personal AI assistant "Cloudbot" (originally Clawd), and founder of the PDF framework PSPDFKit. Peter shares his unconventional current approach to software engineering—shipping large volumes of code he hasn't manually reviewed—enabled by the latest AI agent tooling. The conversation dives deep into how his workflows, team dynamics, and philosophies around code and product building have transformed in the AI era, and why "prompt requests" are supplanting pull requests in his toolkit. Both aspiring engineers and tech leaders will discover actionable insight into the rapidly evolving state of software development.

Key Discussion Points and Insights

Peter’s Journey: From PSPDFKit to Burnout to Cloudbot

[01:12–18:26]

Backstory: Grew up in rural Austria, was a self-taught introvert who loved building things as much as playing games.
Built and scaled PSPDFKit, a PDF rendering framework now used on a billion devices.
Experienced severe burnout as founder and CEO, sold his shares and took a three-year hiatus from tech.
"For a while I was... like, why bother? You're not supposed to retire so early... That messed with my mind quite a bit." (Peter, 32:09)
In early 2025, was drawn back to code with a fresh mindset and a desire to experiment—not for money, but to build “things people find amazing”.

Building PSPDFKit: “Make it like Apple would, with all the love and polish”

[04:39–29:40]

Started selling code "accidentally" by productizing PDF parsing for magazine apps; made thousands from early app sales and reinvested effort into refining the framework.
Cared deeply about user feel, polish, and developer experience, not just features. Hired globally and remotely.
“I love tweaking the details…My product was more successful because developers tried the different ones, and mine just felt the best. I think software is all about how it feels much more than the feature set.” (17:12)
Heavy use of technical blogging as both marketing and team-building; forced engineers to dedicate a day per month to writing public posts.

Burnout, Exit, and Stepping Away

[29:40–33:20]

Burnout was due to the accumulation of stress, nonstop work, and the interpersonal complexity of leadership.
“As a CEO, you’re basically the waste bin because everything that other people don’t manage... you have to fix.”
After selling shares, Peter had “hard years”, catching up on lost time, with months away from his computer.

Return to Tech and Encounter with AI

[33:20–39:01]

Peter first interacts with modern AI code agents (Claude Code, Gemini, Codex) after three years out, approaching them without preconceived skepticism.
“I dragged [my old project] into...Gemini 2.5… I typed 'write me a spec'… and I dragged the spec into Cloud Code and was like 'build'…and eventually it said it's 100% production ready. And I started it and it crashed.” (36:02)
Rapidly recognizes the transformative productivity of AI agents, begins “prompting” as main development method.

New Workflow: “I’m the architect, the agents do the code”

[41:37–56:27]

Core philosophy:
- System architecture and design matter more than reviewing every line of code.
- Most code is “boring plumbing”. The hard, interesting bits require thoughtful architecture, not manual repetition.
Workflow:
- Relies on conversational planning with AI agents (Codex, Claude) to flesh out features and systems.
- Runs multiple agents (5–10) in parallel; “It feels like StarCraft – you have your main base and your side bases.”
- Deeply involved in designing feedback/test loops, uses CLIs over APIs for modular, scriptable integrations.
- “Many other details are boring… It’s much more about system architecture than having to read every single line.” (Peter, 42:58)
Productivity: Routinely ships 600+ commits/day, but only reads code that’s architecturally interesting or complex.

Closing the Loop: Validation, Testing, and Trust

[57:11–63:45]

“The big secret is: you have to close the loop. It needs to be able to debug and test itself.”
Designs systems so that agents can run and validate their own code, improving both test coverage and code quality.
“Using agent-coding makes you a better coder, because you think harder about your architecture so that it’s easier verifiable… verifying is the way to make things good.” (Peter, 61:20)

The Pragmatic Shift: Prompts over Pull Requests

[97:54–102:18]

Code reviews and traditional PRs are “dead” or recontextualized, now called “prompt requests”—focus is on the quality of the prompt and architecture, not code style.
Pull requests from contributors are more valuable for their prompt/goal description than for actual code—because AI rewrites most of it in the new workflow.

Effects on Team Structure & Hiring

[68:02–71:36], [103:05–104:19]

Modern team could operate at 30% the size, with fewer but more senior engineers who are skilled at steering agents.
Hiring profile: curious builders, not code monkeys; open source, prompt-aware collaborators.
“I could easily run a company with 30% of the people. You need really senior engineers who understand what’s important to work on, and which parts you can vibe.” (68:02)
Candid about the “fiasco” this means for tech employment, but insists the shift is inevitable.

The Cloudbot Project: Building “the Future of Siri” by Hacking Together Agents

[73:50–89:12]

Cloudbot: An infinitely extensible personal assistant agent running on user hardware, controlling everything from WhatsApp to smart home devices with natural prompt-based interaction.
Built upon a growing “army of CLIs”, enabling flexibility and agent scripting (as opposed to limited MCP integrations).
Unprecedented viral growth—3000+ GitHub stars in one week.
Focus is on user delight and hiding tech complexity: “The technology disappears. You just talk to a friend on your phone that is infinitely resourceful.”

AI Engineering in Teams: Challenges for Legacy Organizations

[95:19–99:54]

Larger teams/legacy orgs are unlikely to successfully adopt agentic engineering without “refactoring not just your codebase, but your whole company.”
The new role is a “builder” with full-stack product vision, not a manager or a line-coder.
“I design the codebase so it has to be easy for the agent, not for me.” (95:19)

Educational Outlook: New Grads & Learning Advice

[104:45–106:43]

Success for new engineers is about curiosity, experimentation, and learning to ask great questions; traditional university may be poorly equipped to teach this.
“You have an infinitely patient machine that is able to explain you all the things. But it requires real curiosity.”

Notable Quotes & Memorable Moments

On Publishing Unread Code:
“Even now... confession: I ship code I don't read... I care a lot about the structure—did I read all the code? No, because a lot of code is boring plumbing." (Peter, 41:37)
On the Prompt (Not Code) Revolution:
“I read the prompts more than I read the code. To me this is a way higher signal… If someone wants a feature, I ask for a prompt request.” (Peter, 107:54)
On AI Improving Software Quality:
“I write better code now that I don’t write code myself anymore. And I wrote really good code.” (Peter, 62:08)
On Closing the Loop:
“You have to close the loop. The model always needs to be able to verify the work itself, which steers me to better architecture.” (Peter, 63:45)
On Legacy Team Structures:
"You can probably trim the company down to like 30%... The current companies cannot very successfully use AI.” (Peter, 95:19)
On Working with AI Agents:
“It is completely the same feeling for me… flow state, but mentally even more taxing, because I don’t have one employee I manage—I have 5 or 10 that all work on things at once.” (Peter, 54:36–56:27)
On Motivation Post-Exit:
“You’re not supposed to retire so early... That messed with my mind quite a bit.” (Peter, 32:09)
On Team Collaboration Today:
“We don’t talk code. We talk about architecture, big decisions. You still need to have style.” (Peter, 100:05)
On AI-Driven Hiring:
“Someone who's active on GitHub and does open source, and someone where I have the feeling that they love the game... the way you learn in this new world is by trying stuff. It feels like a game.” (Peter, 103:05)

Timestamps for Key Segments

Early Days & PSPDFKit Genesis: 01:12–18:26, 20:38–28:48
On Building for Developers: 18:19–19:13, 26:08–28:48
Tech Burnout and Time Off: 29:40–33:20
Re-entering Tech & First Experiences with AI Agents: 34:33–39:01
Prompt-driven Development and Multi-Agent Workflows: 41:37–56:27
On Validating and Testing AI-Generated Code: 57:11–63:45
Teamwork Changes: Code Reviews and CI/CD: 97:54–102:18
The Philosophy of Modern Developer Teams: 68:02–71:36, 103:05–104:19
Cloudbot's Boom and Future Vision: 73:50–89:12
AI in Big Companies - Adoption Barriers: 95:19–99:54
Advice for New Grads: 104:45–106:43

Final Thoughts

Peter Steinberger embodies the “builder-first” approach that’s emerging as AI agents transform the engineering landscape. His lessons: code is less visible but more testable, prompting and system design are the new core skills, and entire organizations—not just tooling—must evolve. The “prompt request” may soon be as fundamental as the pull request. For engineering leaders and hands-on devs alike: the future belongs to those who can architect systems, shape agents, and—above all—close the loop.

For more insights and the latest on engineering best practices, subscribe to the Pragmatic Engineer newsletter.

The Pragmatic Engineer Podcast

Episode: The Creator of Clawd: "I Ship Code I Don't Read"
Host: Gergely Orosz
Guest: Peter Steinberger (creator of Cloudbot; founder of PSPDFKit)
Date: January 28, 2026

Overview

Key Discussion Points and Insights

Peter’s Journey: From PSPDFKit to Burnout to Cloudbot

[01:12–18:26]

Backstory: Grew up in rural Austria, was a self-taught introvert who loved building things as much as playing games.
Built and scaled PSPDFKit, a PDF rendering framework now used on a billion devices.
Experienced severe burnout as founder and CEO, sold his shares and took a three-year hiatus from tech.
"For a while I was... like, why bother? You're not supposed to retire so early... That messed with my mind quite a bit." (Peter, 32:09)
In early 2025, was drawn back to code with a fresh mindset and a desire to experiment—not for money, but to build “things people find amazing”.

Building PSPDFKit: “Make it like Apple would, with all the love and polish”

[04:39–29:40]

Started selling code "accidentally" by productizing PDF parsing for magazine apps; made thousands from early app sales and reinvested effort into refining the framework.
Cared deeply about user feel, polish, and developer experience, not just features. Hired globally and remotely.
“I love tweaking the details…My product was more successful because developers tried the different ones, and mine just felt the best. I think software is all about how it feels much more than the feature set.” (17:12)
Heavy use of technical blogging as both marketing and team-building; forced engineers to dedicate a day per month to writing public posts.

Burnout, Exit, and Stepping Away

[29:40–33:20]

Burnout was due to the accumulation of stress, nonstop work, and the interpersonal complexity of leadership.
“As a CEO, you’re basically the waste bin because everything that other people don’t manage... you have to fix.”
After selling shares, Peter had “hard years”, catching up on lost time, with months away from his computer.

Return to Tech and Encounter with AI

[33:20–39:01]

Peter first interacts with modern AI code agents (Claude Code, Gemini, Codex) after three years out, approaching them without preconceived skepticism.
“I dragged [my old project] into...Gemini 2.5… I typed 'write me a spec'… and I dragged the spec into Cloud Code and was like 'build'…and eventually it said it's 100% production ready. And I started it and it crashed.” (36:02)
Rapidly recognizes the transformative productivity of AI agents, begins “prompting” as main development method.

New Workflow: “I’m the architect, the agents do the code”

[41:37–56:27]

Core philosophy:
- System architecture and design matter more than reviewing every line of code.
- Most code is “boring plumbing”. The hard, interesting bits require thoughtful architecture, not manual repetition.
Workflow:
- Relies on conversational planning with AI agents (Codex, Claude) to flesh out features and systems.
- Runs multiple agents (5–10) in parallel; “It feels like StarCraft – you have your main base and your side bases.”
- Deeply involved in designing feedback/test loops, uses CLIs over APIs for modular, scriptable integrations.
- “Many other details are boring… It’s much more about system architecture than having to read every single line.” (Peter, 42:58)
Productivity: Routinely ships 600+ commits/day, but only reads code that’s architecturally interesting or complex.

Closing the Loop: Validation, Testing, and Trust

[57:11–63:45]

“The big secret is: you have to close the loop. It needs to be able to debug and test itself.”
Designs systems so that agents can run and validate their own code, improving both test coverage and code quality.
“Using agent-coding makes you a better coder, because you think harder about your architecture so that it’s easier verifiable… verifying is the way to make things good.” (Peter, 61:20)

The Pragmatic Shift: Prompts over Pull Requests

[97:54–102:18]

Code reviews and traditional PRs are “dead” or recontextualized, now called “prompt requests”—focus is on the quality of the prompt and architecture, not code style.
Pull requests from contributors are more valuable for their prompt/goal description than for actual code—because AI rewrites most of it in the new workflow.

Effects on Team Structure & Hiring

[68:02–71:36], [103:05–104:19]

Modern team could operate at 30% the size, with fewer but more senior engineers who are skilled at steering agents.
Hiring profile: curious builders, not code monkeys; open source, prompt-aware collaborators.
“I could easily run a company with 30% of the people. You need really senior engineers who understand what’s important to work on, and which parts you can vibe.” (68:02)
Candid about the “fiasco” this means for tech employment, but insists the shift is inevitable.

The Cloudbot Project: Building “the Future of Siri” by Hacking Together Agents

[73:50–89:12]

Cloudbot: An infinitely extensible personal assistant agent running on user hardware, controlling everything from WhatsApp to smart home devices with natural prompt-based interaction.
Built upon a growing “army of CLIs”, enabling flexibility and agent scripting (as opposed to limited MCP integrations).
Unprecedented viral growth—3000+ GitHub stars in one week.
Focus is on user delight and hiding tech complexity: “The technology disappears. You just talk to a friend on your phone that is infinitely resourceful.”

AI Engineering in Teams: Challenges for Legacy Organizations

[95:19–99:54]

Larger teams/legacy orgs are unlikely to successfully adopt agentic engineering without “refactoring not just your codebase, but your whole company.”
The new role is a “builder” with full-stack product vision, not a manager or a line-coder.
“I design the codebase so it has to be easy for the agent, not for me.” (95:19)

Educational Outlook: New Grads & Learning Advice

[104:45–106:43]

Success for new engineers is about curiosity, experimentation, and learning to ask great questions; traditional university may be poorly equipped to teach this.
“You have an infinitely patient machine that is able to explain you all the things. But it requires real curiosity.”

Notable Quotes & Memorable Moments

On Publishing Unread Code:
“Even now... confession: I ship code I don't read... I care a lot about the structure—did I read all the code? No, because a lot of code is boring plumbing." (Peter, 41:37)
On the Prompt (Not Code) Revolution:
“I read the prompts more than I read the code. To me this is a way higher signal… If someone wants a feature, I ask for a prompt request.” (Peter, 107:54)
On AI Improving Software Quality:
“I write better code now that I don’t write code myself anymore. And I wrote really good code.” (Peter, 62:08)
On Closing the Loop:
“You have to close the loop. The model always needs to be able to verify the work itself, which steers me to better architecture.” (Peter, 63:45)
On Legacy Team Structures:
"You can probably trim the company down to like 30%... The current companies cannot very successfully use AI.” (Peter, 95:19)
On Working with AI Agents:
“It is completely the same feeling for me… flow state, but mentally even more taxing, because I don’t have one employee I manage—I have 5 or 10 that all work on things at once.” (Peter, 54:36–56:27)
On Motivation Post-Exit:
“You’re not supposed to retire so early... That messed with my mind quite a bit.” (Peter, 32:09)
On Team Collaboration Today:
“We don’t talk code. We talk about architecture, big decisions. You still need to have style.” (Peter, 100:05)
On AI-Driven Hiring:
“Someone who's active on GitHub and does open source, and someone where I have the feeling that they love the game... the way you learn in this new world is by trying stuff. It feels like a game.” (Peter, 103:05)

Timestamps for Key Segments

Early Days & PSPDFKit Genesis: 01:12–18:26, 20:38–28:48
On Building for Developers: 18:19–19:13, 26:08–28:48
Tech Burnout and Time Off: 29:40–33:20
Re-entering Tech & First Experiences with AI Agents: 34:33–39:01
Prompt-driven Development and Multi-Agent Workflows: 41:37–56:27
On Validating and Testing AI-Generated Code: 57:11–63:45
Teamwork Changes: Code Reviews and CI/CD: 97:54–102:18
The Philosophy of Modern Developer Teams: 68:02–71:36, 103:05–104:19
Cloudbot's Boom and Future Vision: 73:50–89:12
AI in Big Companies - Adoption Barriers: 95:19–99:54
Advice for New Grads: 104:45–106:43

Final Thoughts

For more insights and the latest on engineering best practices, subscribe to the Pragmatic Engineer newsletter.

wavePod

The creator of Clawd: "I ship code I don't read"

Powered by Wave AI

Summary

The Pragmatic Engineer Podcast

Overview

Key Discussion Points and Insights

Peter’s Journey: From PSPDFKit to Burnout to Cloudbot

Building PSPDFKit: “Make it like Apple would, with all the love and polish”

Burnout, Exit, and Stepping Away

Return to Tech and Encounter with AI

New Workflow: “I’m the architect, the agents do the code”

Closing the Loop: Validation, Testing, and Trust

The Pragmatic Shift: Prompts over Pull Requests

Effects on Team Structure & Hiring

The Cloudbot Project: Building “the Future of Siri” by Hacking Together Agents

AI Engineering in Teams: Challenges for Legacy Organizations

Educational Outlook: New Grads & Learning Advice

Notable Quotes & Memorable Moments

Timestamps for Key Segments

Final Thoughts

Summary

The Pragmatic Engineer Podcast

Overview

Key Discussion Points and Insights

Peter’s Journey: From PSPDFKit to Burnout to Cloudbot

Building PSPDFKit: “Make it like Apple would, with all the love and polish”

Burnout, Exit, and Stepping Away

Return to Tech and Encounter with AI

New Workflow: “I’m the architect, the agents do the code”

Closing the Loop: Validation, Testing, and Trust

The Pragmatic Shift: Prompts over Pull Requests

Effects on Team Structure & Hiring

The Cloudbot Project: Building “the Future of Siri” by Hacking Together Agents

AI Engineering in Teams: Challenges for Legacy Organizations

Educational Outlook: New Grads & Learning Advice

Notable Quotes & Memorable Moments

Timestamps for Key Segments

Final Thoughts