The Unreasonable Effectiveness of Code

These tools are about way more than just "code." Or equivalently: we vastly underestimate the versatility of code.

Physicist Eugene Wigner wrote about "the unreasonable effectiveness of mathematics" — how math explains so much of the physical universe beyond what it was designed to describe. Code plus the right LLM harness is unreasonably effective at knowledge work in the same way. Very few people understand this yet.

Here's a random example. At a hunch, I threw raw astrophotography data — a FITS file — at VerbaGPT and asked it to adjust the image to bring out the Orion Nebula and the details within. This is something that normally takes hours of work with highly specialized software (PixInsight, Siril, etc.) and requires significant image-processing skill. I have neither.

VerbaGPT found an appropriate Python package, applied dozens of image-processing edits — color stretching, contrast adjustment, star-masking, and other techniques I wouldn't have known to apply — and produced this:

That's the Orion Nebula from my driveway. About 10 minutes of exposure time. A Bortle 6 sky (moderately light-polluted suburban area). And 15 minutes of VerbaGPT doing what usually takes hours of manual post-processing.

The Broader Point

The obvious application for VerbaGPT is data analysis. But the underlying capability — natural language → Python code → execution → result — applies anywhere Python does. Image processing, file conversion, statistical modeling, document generation, scientific computation.

The constraint has never been whether Python can do something. Python can do almost anything. The constraint has been whether you know how to write the Python. That constraint is dissolving.

What tasks in your workflow require specialized skills you don't have? That's worth thinking about.

Originally posted on LinkedIn · January 15, 2026