The big theme of software engineering in 2026 is very clearly about AI coding agents. We’ve solidly passed the point asking if they can code well. They can. Now we’re all figuring out “okay, they can code well, now what?” This is one of the things I keep coming back to when I think through that question.
(Between the time I first had this thought to when I’m actually writing this post, code is cheap feels like a pretty common position now. But hey, it’s my blog and I’ll publish if I want to.)
For my entire career as a software engineer, code has been expensive. I mostly mean this in a “time is money” sense, but obviously tech industry salaries clearly show: engineers ain’t cheap. But just in terms of human working hours required, coding was an artisanal craft. Working with any significant amount of code needed a lot of time for an engineer to get to continously understand it, reason about it, and then modify and maintain it. All of that time cost a lot of money, and that’s just been a foundational truth of how the industry built everything.
With the AI coding agents lately, it feels like that’s no longer feels true. In minutes, Claude can digest a massive codebase and spit out worthwhile changes for just a few dollars. All of the time and money calculus of working with code up to this point has permanently changed. Code is cheap now.
Cheap code is mass-produced in bulk
When code is cheap, we make more of it.
AI coding agents have measurably increased the lines of code written across companies and FOSS projects. They write code way faster than people, and people are using them to write code that they wouldn’t have otherwise written. In my experience too, AI coding agents are great at idea expansion but less naturally interested in idea consolidation. They need extra guidance to refactor and abstract things. And it’s worth calling out that mass-produced doesn’t necessarily imply worse quality either. They mass produce unit tests and documentation too; after all those things are code too, and code is cheap now.
So the total volume of code in the world is going up. More code overall also means the ability of individual humans to understand whole codebases is likely going down. So we’ll need coding agents even more than now to work with our codebases growing ever larger due to coding agents. The cycle repeats.
Cheap code upends timeline estimates
Writing 1000 lines of decent code used to take a week. Coding agents can make it take a few minutes.
This turns out new feature ideas quickly, obviously. We’ve all seen our fair share of backlogs full of dead features that might be nice to have but are not even worth the time to prototype. But code is cheap now, prototyping and scaffolding are trivial tasks.
But it also enables new possibilities with more engineer-facing work like of refactoring and automation scripting. This is a blessing. I’ll never again be forced to waste my life learning how to use yet another “ultra-powerful” DSL for some random utility tool. Working in legacy codebases is a whole new ballgame. You can write a script that plugs together a couple of databases and APIs so quickly now that creating automations that previously were not worth the time can actually be faster than doing them manually. On that xkcd automation chart chart, they’ve meaningfully moved up and to the left. It’d be malpractice not to automate some things now, Larry Wall would say you’re not being a virtuous programmer.
And it’s a lot easier saying no to something that doesn’t exist than saying no to something that does. Easier prototyping can lead to more features. But more features is not necessarily a better end product. And the more your end product looks not quite fit for simple problems, the more people will look elsewhere to solve their simple problems. Which leads naturally to the next point.
Cheap code crowds existing niches and finds new ones
Everybody, their mom, and especially their boss can all code a simple application now.
Lots of apps fundamentally are simple but would be annoying to implement yourself. Those are now in danger of being crowded out by a bunch of new custom variations on the theme. Things where the user only wants 20% of the features for 1% of the price. Or things where the user wishes you would just create this one little feature that you’re reluctant to do. Or where the industry standard is crap but it’s still the standard and nobody else wants to even try replacing it. Tons of software engineers I know have said something to the effect of “if I just had time, I could bring back Google Reader.” Well, code is cheap now, get on it.
The same dynamic also works for super niche app ideas that otherwise likely wouldn’t get much interest. A great example of this is Joe Weisenthal’s purely vibecoded app havelock.ai, which judges a given body of text on its orality. No, I had never heard of that concept until I saw that website either.. There are almost definitely more niche apps that nobody ever expected to be worth coding up that look a lot more reasonable.
But I think a lot of the most interesting stuff here will all be the semi-technical professionals that now can have their own personal intern for coding things. There are a tons of jobs where an SME knows exactly what they want the computer to do, but they personally don’t know how to make the computer do that. If they’re extremely lucky, they have access to a department full of good and available engineers who will respond promptly to their requests. But hahaha, come on, those don’t exist. Instead most of those ideas end up just never getting executed at all. But these days folks could likely get a decent proof of concept working all by themselves, certainly they can end up with something way better than they’d have otherwise.
Cheap code means cheaper coders?
This is still unclear to me. For sure it feels likely. It’s definitely a narrative driving a lot of the industry’s layoffs and job trends lately. But I’m not sure it’s guaranteed. Jevons paradox suggests that when code is cheaper to produce, there will be more demand for it. Does that mean similar wages but fewer jobs, more jobs with worse wages, or some other equilibrium? Unclear.
But of course the hardest parts of software engineering were never really the coding. The hardest parts were about knowing what to code and why, and how to bring along all of the stakeholders and users with you on that journey. That’s why every career ladder in existence for software engineers emphasizes those skills more and more as you move up to more senior levels. AI agents are great coders. But I’m not sure they’re great at the people-to-people work of building trust to answer those top level questions. Maybe I’m a romantic on that one.
Still, one thing that definitely feels true is that if you’re a software engineer, good code can no longer be the primary value you bring to your job. Code is cheap now.