How One Developer Merges 600 Commits a Day with AI: Inside the Clawdbot Workflow

Background

The Pragmatic Engineer podcast invited Peter Steinberger, founder of PSPDFKit and creator of the AI‑driven assistant Clawdbot (now Moltbot), to discuss the radical changes AI brings to software development.

Peter Steinberger’s Journey

Peter grew up in rural Austria, taught himself programming, built early iOS apps that earned his first revenue, and later founded PSPDFKit, a PDF rendering framework used on over a billion devices. After selling the company and experiencing burnout, he returned to development in 2023, drawn by the capabilities of Claude Code and Anthropic’s O1 models.

Core Interview Insights

1. “Human Merge Button” – 600 Commits a Day

Peter disclosed that he can complete around 600 commits in a single day, most of them generated by AI agents. Instead of writing each line, he reviews and merges AI‑produced changes, acting as a “human merge button.”

2. Closing the Loop

His most controversial methodology is the “closing the loop” principle: he ships code he has never read because the AI writes tests, runs the CLI, lints, and only if all automated gates pass does he merge. This validation loop makes AI‑generated code trustworthy.

3. Prompt Request Replaces Pull Request

Peter argues that traditional pull requests are dead. He now focuses on the prompt that generated the change, treating it as a “Prompt Request.” By adjusting the prompt, the model can regenerate the code without manual line‑by‑line edits.

4. Multi‑Agent “Chess”

Clawdbot runs 5‑10 parallel agents that handle different tasks—coding, documentation, bug fixing—allowing Peter to stay in a state of flow while the agents occupy the idle seconds between model calls.

5. Tooling Choices: CLI vs. MCP

Peter prefers command‑line tools (e.g., jq, grep) over Model‑Centered Platforms (MCP) because they let agents filter data without loading massive tool definitions into the model’s context.

Impact on Software Engineering

The interview highlights how AI changes the economics of development: fewer engineers can produce more output, but the role shifts toward architectural oversight, prompt engineering, and system design. Traditional code‑review culture fades, while verification through automated tests and linting becomes the primary quality gate.

Advice for Developers

Develop strong system‑understanding skills and curiosity.

Read complex open‑source code to learn patterns that AI can emulate.

Invest time in writing good prompts; they become the main communication with the model.

Use AI to generate documentation and tests, not just code.

Future Outlook

Peter predicts that as models become more reliable, the distinction between developer and AI blurs. Companies will need “high‑agency” engineers who can design workflows that let AI handle routine implementation while they focus on high‑level product vision.