Pean preserves the meaning and organization of content. Every response is grounded in the original source, and you can explore, question, and reuse knowledge naturally.

How does Pean help me understand content?

Pean transforms long-form content into structured, navigable knowledge. You can see how ideas are organized and connected, ask meaningful questions with grounded answers, and reuse knowledge confidently.

Pean uses advanced AI technology to analyze YouTube videos. We first transcribe the video using AI speech recognition models, then use large language models to extract structure, generate summaries, create chapters, and build mind maps. The entire process takes just seconds.

What types of videos work best?

Pean works with any YouTube video that has audio. Educational content, tutorials, lectures, podcasts, and presentations work particularly well. The clearer the audio, the better the transcription and understanding quality.

Can I translate summaries to other languages?

Absolutely! Pean supports multiple languages. You can translate both summaries and transcripts into your preferred language instantly using our language switcher.

Is my data private and secure?

Yes, we take privacy seriously. We only process the video content you submit and don't share your data with third parties. Video summaries are stored securely and associated with your account for your reference.

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

Overview

Peter Steinberger, the creator of PSPDFKit and the new project Clawbot, discusses his unconventional approach to software development in the AI era. After a successful exit and a multi-year hiatus, he returned to find AI coding agents transformative. He now builds software by architecting systems and prompting AI agents to generate and weave in code, often without reading the generated lines. He argues that traditional practices like detailed code reviews are becoming obsolete, replaced by a focus on system design, closing the feedback loop for validation, and treating pull requests more as prompt requests. His current project, Clawbot, is a hyper-personal assistant that feels like a glimpse into the future of human-computer interaction.

Timeline Summary

🐝 Early Career and PSPDFKit

Peter got into tech in rural Austria after being inspired by a summer guest's computer, leading him to tinker with DOS games and early web development.
His first major project was an iOS app for a dating website, built by parsing HTML, which he sold for $5 and made significant early revenue.
The idea for PSPDFKit emerged from helping a friend with a poorly coded magazine viewer app; he rewrote the PDF component, which later became the foundation of his billion-device framework.
He started selling the PDF component directly via a simple website, and the business grew through word-of-mouth and a developer-focused marketing strategy.

🔥 Burnout, Break, and AI Rediscovery

After 13 years, Peter experienced severe burnout from the relentless pace and managerial burdens of running PSPDFKit, leading him to sell his shares and take a 3-year break from tech.
He returned in April 2024 with a fresh perspective and immediately explored the new AI coding tools, starting with Claude Code.
His initial experiments, like having an AI build a spec from a massive code dump, were flawed but revealed the technology's potential, hooking him completely.
This period reignited his passion, leading to sleepless nights and what he jokingly called the "black eye club" with friends equally addicted to exploring AI's capabilities.

🤖 Building Clawbot and a New Workflow

Clawbot originated from a desire to build a deeply personal, proactive assistant, inspired by the film "Her," and started as a simple WhatsApp relay for remote computer control.
The project exploded in popularity after he added it to a public Discord, allowing others to see its capabilities, such as controlling smart home devices and acting as a DJ.
Peter's development workflow involves managing multiple AI agents in parallel, designing system architecture through conversation, and having the agents "weave" code into the existing structure.
He emphasizes "closing the loop" as the key principle, ensuring agents can compile, test, and validate their own work, which makes AI exceptionally effective for coding.

Key Points

🚀 AI Transforms the Builder's Role:Peter describes his current role as a "builder" or architect who designs systems through conversation with AI agents, focusing on outcomes and product feel rather than line-by-line coding.
🔄 The Critical "Closing the Loop" Principle:The secret to effective AI-assisted development is designing systems so the AI can validate its own work through compilation, testing, and execution, creating a tight feedback loop.
📈 Prompts Over Pull Requests:In the AI era, the thinking and prompting that lead to a solution are more valuable than the final code. Peter prefers "prompt requests" over pull requests and often rewrites contributions based on the original intent.
🎯 Product Polish and Developer Experience Mattered:PSPDFKit's success was attributed not to feature count but to its polished feel and excellent developer experience, a philosophy he carries into his AI-driven projects.
💼 Enterprise Sales Involves Custom Pricing:He explains that enterprise software often uses custom pricing ("call us") because value and usage vary dramatically between a freelancer and a Fortune 500 company, making a single price point ineffective.
😓 Burnout Stemmed from Managerial Load, Not Work:His burnout resulted from the loneliness of leadership, constant firefighting, and people drama, not from the act of coding or building itself.

Frequently Asked Questions (FAQs)

Why does Peter "ship code he doesn't read"?
He believes much of modern app code is "boring plumbing" for data transformation; he focuses on system architecture and trusts AI agents to generate the correct, validated boilerplate code.
What is "closing the loop" and why is it important?
It means designing the development process so AI agents can automatically compile, lint, run tests, and verify their output. This feedback loop is why AI is particularly good at coding compared to creative tasks.
Why are traditional code reviews becoming less useful?
With AI, the valuable artifact is the prompt and the architectural thinking behind it. Reviewing the generated code is often less efficient than reviewing the prompt and having an agent implement the vision directly.
How does Peter's workflow with AI agents actually work?
He runs multiple agents (like OpenAI's Codex) in parallel, having conversations to plan features, then instructs them to "weave" code into the codebase. He juggles these parallel tasks like a strategist managing multiple game boards.
What are the biggest challenges for companies adopting AI effectively?
It requires a complete redefinition of roles and processes, favoring high-agency, product-focused builders over specialized roles. Existing codebases and company structures are significant barriers to this change.
What advice does he have for new developers entering the field?
Be infinitely curious, use AI as a patient tutor to understand complex systems, and focus on building things to gain practical experience, even if you're not writing every line of code yourself.

Conclusion

Peter Steinberger's journey highlights a seismic shift in software development, where AI agents handle implementation details, allowing developers to focus on architecture, product vision, and user experience. His experience building PSPDFKit provided the foundational knowledge that now allows him to effectively guide AI. While this new paradigm is highly productive for individual builders and small teams, it poses significant adaptation challenges for larger organizations with entrenched processes. The core skills of the future may revolve around system design, taste, and the ability to communicate effectively with AI.

Action Suggestion: Experiment with setting up a "closed loop" for an AI agent on a small project, where it must write, run, and pass tests for its own code before considering the task complete.