Bas Nijholt's Agentic Coding Journey

Bas Nijholt is a Staff Engineer at IonQ building quantum computers, with a PhD in computational quantum mechanics from TU Delft. He’s an open-source enthusiast who created Adaptive, a Python package for parallel parameter sweeps that’s been downloaded over 300,000 times and saved researchers millions of hours of compute time. Before IonQ, he worked at Microsoft Quantum on topological quantum computing tools.

Nijholt went from declaring he’d “never invest in, build upon, or use” AI coding products to shipping more Python packages than ever before. His November 2025 blog post documents the transformation.

The Skeptic Phase

In early 2025, Nijholt had a disaster with “vibe coding” — letting AI generate code without careful review:

“What started as a fun 3-hour prototype turned into a 15-hour debugging nightmare.”

He concluded AI was dangerous without human oversight and publicly swore off AI coding tools.

What Changed

Eight months later, his view shifted — not his core principle (he still never merges code he doesn’t understand), but the tooling improved. Two key developments:

1. Agents can run tests. When AI can execute your test suite and iterate on failures, the feedback loop closes automatically.
2. Better context handling. Modern tools maintain project context across sessions.

The Evidence

Nijholt tracks his productivity through PyPI package publications:

The graph in his blog post shows a clear inflection point after adopting agentic tools.

Background

Nijholt holds a PhD in computational quantum mechanics from TU Delft, where he researched semiconductor-superconductor hybrid nanowire structures that can host Majorana bound states. Before IonQ, he worked at Microsoft Quantum building tools for topological quantum computers.

His open source contributions include:

• Adaptive — Parallel parameter sweep optimization, used by researchers worldwide
• Home Assistant Stream Deck YAML — Control home automation via Stream Deck
• Multiple Python packages for scientific computing and automation

He documents his tooling preferences at nijho.lt, including posts on Python development tooling and file-based RAG without databases.

Core Principle: Never Merge Code You Don’t Understand

This rule survived his skeptic-to-advocate transition:

“I still never merge code I don’t understand — but the tooling and my approach have evolved.”

The difference is how that understanding happens:

• Before: Write every line yourself
• Now: Review and understand AI-generated code before committing

Home Automation as Test Case

Nijholt’s Home Assistant Stream Deck project became an early AI-assisted development experiment. The project lets users control temperature, lights, TV, and music via programmable Stream Deck buttons.

With 250+ GitHub stars and 9 contributors, it grew beyond his initial vision — partly because AI helped him iterate faster on community feature requests.

Tool Stack

Based on his blog posts and LinkedIn activity:

Key Insights

From the agentic coding post:

Let agents run tests. The breakthrough came when AI could execute tests and fix failures autonomously.

Track your output. Nijholt measures productivity through concrete artifacts (published packages), not subjective feelings.

Maintain oversight. AI accelerates iteration, but humans must understand every change before merging.

Takeaways

Links

• Blog: On Agentic Coding
• Blog: Parallel Agentic Coding
• Blog: File-Based RAG & Memory
• Blog: Python Dev Tooling
• GitHub: @basnijholt
• Adaptive Documentation
• Google Scholar
• LinkedIn: Bas Nijholt

Next: Alexander Opalic’s Claude Code Deep Dives