In the future, everyone will be an engineer - but is that good?
Some thoughts on how AI's impact on software engineering will play out, and what we can learn from other industries on what that means for all of us in software.
It’s hard to imagine many topics getting more ink on paper over the last year than the societal impact of artificial intelligence. As an engineer, an area of special interest to me is the impact of AI on engineering - software engineering in particular. I think so far, coding is probably one of the 3-4 areas where AI has had the most substantial impact, and also where there is a lot of money going to solving very real and tractable problems: Co-pilots, no code environments, UI automation, testing tools and so on. But we’re still just a few years in on this journey, and to quote Roy Amara: “We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”
So with that in mind – what predictions can we make about the long term impact of AI on software engineering? I think there are two: Even more of the world will be eaten by software, and power law dynamics will shape the financial outcomes of this.
These are, I think, not particularly unique or novel thoughts. Marc Andreesen coined that software is eating the world 13 years ago, and on power laws, Chris Anderson wrote the Long tail in 2006, and more recently Ben Thompson’s Stratechery is discussing power laws on an almost weekly basis. So I’m not writing this down with the aspiration to contribute something new, as much as to get my own thoughts in order. With caveats out of the way:
Prediction 1: Even more of the world will be eaten by software
Over the last 20 years, there’s been both an explosion of software and an explosion of software engineers. The sheer amount of stuff in the world now that runs on code is staggering. Banks, phones, doorbells, grocery shops, cars, dating, eating food, education, sex, and so on. If there’s a thing, chances are there is software for that thing.
But – there is still lots more to digitize and create software for. More niches, markets, problems. And I tend to think the rate-limiting step to doing that is the cost and availability of software engineers. Engineers are still costly. That puts a barrier to problems that can’t be solved by software. Areas that aren’t super profitable, super scaled, or that are the pet peeve of someone who can code, likely won’t get digitized.
With AI, this’ll change. Most people will be able to express themselves in code by communicating with an AI. And the infrastructure (i.e. cloud, integrations & apis etc) that will make that possible has been built and matured for 15 years now. So it doesn’t seem unreasonable that more people than ever before with lower cost than ever before can turn things into software, and use software as a tool to solve their problems. My mental model for this is the amount of auto repair shops you see everywhere – writing software being as common as fixing your car.
In terms of the impact for engineers, I’m a bull. I think this explosion of software is good news for engineers, at least as far as stable long term employment goes. Software is easier to create than to maintain. And a future wholly digitized will be a future with a lot of maintenance. Clearly this will be amenable to AI too, but I am less bullish that generic “fix all problems in my code” AI will be universally tractable.
Prediction 2: Power law dynamics will shape financial outcomes
At its core, technological innovation removes constraints, and the removal of those constraints reshape economic outcomes. The internet drops distribution costs to essentially zero, and all of a sudden, newspapers, the music industry and so on all look entirely different than they did in a physical world. If the constraint AI removes for software is the need to have engineers for a lot of work, then that will reshape the economic outcomes for engineers. And the way it’s happened in every other industry is through power laws: An enormous amount of value accumulates to the top of the value chain, and a lot also disperses into the tail, but the middle gets squeezed out.
For software, I think this means the following: There will be a top of the pyramid made up of 100’s or maybe 1000’s of companies that will be incredibly valuable and that will be greater and greater in scale. Think FAANG metastasising by orders of magnitude again. And then there will be millions of people creating software as individuals and creating a good life for themselves at a small scale. But in the middle, there will be less and less economic value to capture. This means mid-tier software shops, consultancies, and so on will compete for an ever shrinking pie.
The counterforce to this is regulation. Both in terms of attempts to limit the power accruing to the top via antitrust, and also in terms of regulatory capture (governments allocating their capital to vendors via RFP processes and so on to keep them alive). Given point 1 – that all of the world will be software, the stakes for this are enormous. And so I’d expect regulation to play a major role in slowing down or limiting what’ll happen. But ultimately, power laws tend to be very hard to stop.
What’s the time horizon?
Finally - how long will this take? Shrug emoji. But summoning Amara again, my best guess is that in 7-10 years, the world will have been reshaped, and everyone and their parents will be coding, even if they don’t even know that that’s what they’re doing. And I think as a result, there will be a handful of immensely valuable software companies (as today, but more so), a few hundred almost as valuable software companies, a big void in the middle, and hundreds of thousands of small, AI-powered firms.