Since you’re reading this you’ve probably already tried an AI tool in your code editor. Whether it’s Copilot, Windsurf, or Cursor, AI-first tools are making huge waves across the industry.
The last couple of weeks were busy for AI in software development. Two big things happened:
OpenAI acquired Windsurf for a rumored $3B
Cursor raised $900M and a $9B valuation
These are some of the fastest growing B2B products of all time so it’s no surprise there’s tons of money to be made.
Every AI lab is obsessed with AI writing code
I think about this a lot.
Software engineering seems to be the industry with the biggest push to use AI on the job. Almost every frontier model is measured on its ability to write code, and there are constantly new benchmarks. There’s a few guesses I might have:
It’s easier to train models to write code than do other types of knowledge work
Software engineers are well-paid so there’s lots of costs savings to be had
The people making these models and products are software engineers (or close enough), so they’re interested in the field
The people doing the research believe they can do the research faster if they automate software engineering first
The first guess here has little basis in reality, rather it feels intuitive. I’ve not see any researcher saying coding is the easiest knowledge work to train models around.
The second guess here is probably my favorite. Software engineers are expensive, so you can charge a lot of money for tools that augment (or replace) them. Think about it - what would a company be willing to pay for a tool that makes their developers twice as effective? What about ten times as effective?
The third guess, that researchers are just doing what they’re interested in, makes sense but isn’t as compelling as financial motivations.
The last guess comes from Mark Zuckerberg! Zuck was on Dwarkesh podcast last week and he made reference to the idea that if we can augment engineers and researchers well enough, their pace of breakthroughs will accelerate. Compelling, but it’s a heck of a bet for a company like OpenAI to make.
Watch what they do, not what they say
Pretty much every AI lab and software engineering tool has a leader publicly saying software engineering is going away. The timelines vary, from 6 months to a decade, but they’re all saying the same thing.
If software engineering is going to be automated, why waste money on tools for software engineers? OpenAI wouldn’t spend billions on a VS code fork that makes API calls to its own models if it thought you could just prompt your way through everything. If anything, they’d have bought something like Lovable or Bolt that has a stronger emphasis on being hands-off-keyboard.
Over the last year I’ve developed the opinion that AI isn’t going to automate software engineering, but it will change it.
It will change it enough that it might be unrecognizable, but the labor market will still need skilled humans that understand they systems they’re working with.
What to do about it
It’s obvious that AI is going to play a big part in the future of software engineering. My belief is based mostly on Cal Newport’s “So Good They Can’t Ignore You”
The labor market has always favored those who bring value that far exceeds their cost. AI tools are going to make the gap between the best engineers and the worst even wider than it is today.
Want to be part of the group providing outsized value? I think it's becoming increasingly hard to do so if you refuse to augment yourself with AI.
That’s what this newsletter is all about.
Great write up bro! I think about it a lot, too. AI has easily 10x my output. It’s not taking my job anytime soon, but people just starting out are going to have a hell of a time bc they’ll be too dependent on it.