This very publication, Daily Maverick, pays me money to do software development that I don’t do. The software is still being produced, which has thus far avoided uncomfortable conversations about my continued employment, but it’s not me doing the work; well, not all of it anyway. 

Instead, my little artificial intelligence buddy, Github Copilot, is at the coalface, banging out lines of code while I supervise, add a bit of guidance here and there, and count my cash.

(I exaggerate a little for effect — I think Copilot is writing about a third to a half of my code, depending on the task. Most of the problems we are trying to solve in code have been solved before, and being the biggest code repository in the world, Github has that solution at hand.) 

The promises of AI have been as varied as the prophecies regarding what will one day be our digital overlords: they’re meant to drive cars, replace all lawyers and accountants, fly autonomous drones and identify enemy combatants for slaughter. They will either create ultimate freedom from labour for humankind, destroy the economy and make us all unemployed, or simply transform the planet and everything on it into microchips, depending on which sci-fi book or futurist you listen to.

In practice, however, they’re still quite rubbish at driving, and sadly have completely failed to eradicate either lawyers or accountants. Quite frankly, AI was starting to smell as fishy as Bitcoin, the Metaverse, and Brexit custom checks — a shared delusion that technology was capable of much more than it truly is.

Then I met my buddy Copilot. I popped him onto my code editor, Visual Studio Code, a development environment created by Microsoft, which, uncoincidentally, bought Github for $7.5-billion in 2018. That’s billion with a B. At the time I was coding in PHP, not what I use every day, so I was a little rusty. I figured this new-fangled coding AI might remind me to put in some missing semicolons, which PHP is obsessed with. If you can remember to put a semicolon at the end of every line, you already know most of what you need to know to code in PHP.

It did not remind me to put semicolons at the end of the line. Instead, I wrote a function name, hit enter, and it produced all the code that the function should do. Instantly. My jaw hit the floor.

I tried another function that I needed. Again, it filled in all the nitty-gritty. Then, I tried writing a plain-English comment of what I wanted my code to do. It wrote the code. After a while, it started writing the comments too.

To explain why I found this so incredible, let me digress a little, to 18th-century Eastern Europe. Hungarian civil servant, polymath and part-time inventor, Wolfgang von Kempelen, presented the Austrian Empress Maria Theresa with an automated chess set that can play a pretty good game of chess against a human. Named The Turk, as it prominently featured a full-sized model of a “Turkish Sorcerer”, this machine toured Europe for 84 years, defeating Napoleon and Benjamin Franklin, among many others. Bear in mind that we didn’t have chess computers until the 1950s, and they were rubbish until the 80s. It was truly a marvel of its time. And, of course, a complete sham. Hidden cleverly in the mechanism was a very cramped chess master, who would move the pieces from underneath the board via magnet. 

Until Copilot, most AI that I engaged with worked a bit like this — either it was useless, or a cleverly written traditional system that used a tower of logic and processing power to fake AI, or literally humans pretending to be machines, supplying answers to questions over the internet. Amazon even has a product called Mechanical Turk (after our 1700s chess-playing contraption), which lets you send computer-like tasks for humans to perform for meagre returns. 

That’s why I was both so impressed with and suspicious of Copilot, and to be honest I was left wondering where they were hiding the tiny chess master. 

So what does this mean for accountants, lawyers, fighter-jet pilots, and of course software developers? Are we all going to lose our jobs and turn to universal income grants for survival? Copilot, I believe (and hope), shows us another path: man and machine working together, to be much more than either can be on our own. If you think that a software developer’s job is to produce code, then Copilot is definitely a threat to software developers. If you think a software developer’s job is to produce solutions to problems, then it is an incredible leap forward. I can free my mind from the minutiae of syntax and semicolons to think on a new level, that of creating solutions to problems. I’m coding twice as fast, and the stuff I put out is significantly better. And I believe the same will be true for other industries that AI is taking aim at: it’s a tool, not an existential threat. 

Copilot has been in a closed Beta since the beginning of the year, but on 21 June 2022 they opened the doors and are taking subscriptions. It costs $10 a month, and honestly I can’t give them my money fast enough. Because, really, it’s not my money, it’s Daily Maverick’s. DM