Essays on AI, software and the shape of technical work, by Cesaire Tobias.
by Cesaire Tobias
For as long as I’ve been coding, my preference has always been towards open-source tools. When I started out, more than 10 years ago — long before ChatGPT, GitHub Copilot, or any AI copilots existed — open source just felt like the natural choice.
Development happens in freely, unencumbered by corporate roadmaps. That means you can access tools at the bleeding edge, while still relying on older, stable packages.
Trust is earned not assumed. If a library like ggplot2 works brilliantly, it becomes a standard because thousands of users independently choose to use it — not because a vendor told you it was “best in class”.
Open source is accessible. As a student or employee without a company credit card, I didn’t need special permission or budget approval to experiment, build, and innovate. I could just download Python or R and get to work.
And most importantly, open ecosystems are unbounded. You’re not locked into a walled garden; you can extend, remix, and connect your tools however you like.
Meanwhile, proprietary software like Stata, SAS, or MATLAB thrived for decades because they solved a different problem: they made advanced analysis accessible to non-programmers. Their value proposition was simple: you didn’t need to be a software engineer to get reliable, reproducible results. Universities taught them. Governments and firms standardized on them. They became safe bets with someone to blame if anything went wrong.
So what’s changed?
The biggest selling point of proprietary languages was their “easy syntax.” For a social scientist, economist, or engineer who wasn’t trained in computer science, it was easier to type a Stata command than to write Python code.
Today, that barrier no longer exists. With AI, you can ask in plain English:
“Run a fixed effects panel regression in Python and plot the residuals.”
…and get working, reproducible, open-source code in seconds.
In other words, AI has flattened the learning curve. The unique selling point that once justified sticking with Stata or MATLAB — “approachability for non-programmers” — no longer holds up when AI makes Python or R just as accessible.
It’s true: proprietary vendors are not ignoring this shift.
MathWorks, for example, already offers the MATLAB AI Chat Playground and promotes a MATLAB Copilot to help users turn natural language into MATLAB-ready code. Other vendors will almost certainly follow suit.
This matters, because it shows proprietary languages aren’t standing still. They are embedding AI directly into their environments, doubling down on their promise of “no setup, no hassle — just results.”
However, once AI has leveled the ease-of-use playing field, the remaining differentiators all favor open source:
So yes, proprietary copilots exist — but they don’t solve the underlying disadvantages.
Maybe not tomorrow. Stata, SAS, and MATLAB still have niches where they’ll survive — in certain research departments, government workflows, and among long-time users who value stability.
But AI has fundamentally undermined their core value proposition. And when you add that to the long-standing advantages of open source — community power, earned trust, accessibility, and freedom — it’s hard to see a future where proprietary analysis languages remain relevant at scale.
I loved open source before AI, because it enabled unencumbered innovation. Now, with AI copilots lowering the barriers to entry, I believe the migration away from proprietary languages is not just inevitable — it’s accelerating.
September 11, 2025
tags: languages - ai - open-source