We recently started using GitHub to manage our code. With this cloud-based tool, multiple developers can work on the same project at once. When their changes conflict, GitHub manages the branches and lets the work be merged back into a single production codebase.
GitHub is owned by Microsoft and is free for developers worldwide. Microsoft’s reputation gives programmers confidence that their sensitive, proprietary work is safe on the platform.
The question is where this leads. Looking into the crystal ball, this is what I see.
The best minds in the world have trusted their most valuable ideas and creations to GitHub. If GitHub eventually offers a code optimisation tool, we will all sign up and accept the productivity gains. Using machine learning trained on the best code in every discipline and language, GitHub will be able to offer bug detection and code optimisation to everyone. Think of it as an advanced spellchecker for code: it will flag inefficient code and suggest better approaches, all learned from the world’s best existing work.
This would be a revolution. You would only need to sketch the rough framework of a project, and the AI would fill in the details as efficiently as possible. Taken further, the AI would build libraries of procedures so efficient that humans could no longer read them. At that point the fly-by-wire concept applies to software itself: humans do the bare minimum of coding, and the AI writes the actual code to a level of precision too complex for any person to follow.
Everyone will be able to write an app or a program, because in effect nobody will be writing code by hand anymore. Working the way we do today would be like building a modern car with a caveman’s hand tools. It’s possible, but it’s a waste of effort.
This points to a future where our whole society runs on systems too complex for anyone to fully understand. As inefficient, human-readable code is phased out, software will advance rapidly and feed a new generation of better AI, in a loop of exponential growth.
It also creates obvious openings for bad actors and hostile states engaged in digital warfare. Left unchecked, sabotage and cyber-terrorism could become the next pandemic.
Until now, AI has been trained mainly on images, video and voice, which are abundant online and which AI has largely mastered. GitHub is a new kind of training source: the collective knowledge of the world’s best computer scientists. I believe the machine learning opportunities that come out of it will be a genuine game changer, in an era already used to game changers.
I could look further into the future, but each step branches into too many possibilities to see the endgame clearly from here. The point stands: GitHub’s decision to open itself to developers for free, and the way we have all flocked to it, will have a significant influence on the path humanity takes.
Postscript, 2021
This article was written before GitHub launched its “Copilot” feature. Now I feel like an oracle.
Looking back, November 2026
Five and a half years on, most of this has happened. Describing what you want and letting AI write the code is now how software gets built, at my own company as much as anywhere. The “advanced spellchecker” arrived within weeks of this post and has grown into something that writes entire systems from a rough brief. Building an app no longer requires knowing how to code.
Two things have not happened, at least not yet. AI-written code is still readable by humans, and we still insist on reading it. The mechanism was also different from what I imagined: Copilot learned from public code, not from the private work people trusted to GitHub. I got the outcome right but the route slightly wrong.
The warning about sabotage and cyber-warfare still stands, and it matters more now than it did then.
M Parak
