What Should We Do About Technical Debt?

I’ve been very fortunate to have a good friend, Bill Chmura, to talk through all of the changes that AI is causing in our society. He has a great perspective with decades of experience as a working coder and manager, and his thinking always pushes my own into areas I hadn’t considered. He publishes at his site, Dispatches from the Far Reaches, and today we’re trying something different: a shared dialog.
Tom
Author and technologist Cory Doctorow says that “AI is the asbestos we’re shoveling into the walls of our civilization and our descendants will be digging it out for generations.” I’m also worried about the technical debt — the cost of maintaining software over time — caused by having humans choose to step back and let AI create code without careful oversight. I’ve been very careful not to distribute or sell any of the software products I’ve created with AI assistance, largely because if I later find out that something’s wrong, I wouldn’t necessarily know how to fix it. How does an experienced coder feel about using AI to create and deploy software for business?
Bill
The fairest answer is I would not be comfortable having a single human developer or an AI create and deploy software on their own. I trust that the human would not knowingly release something terribly broken. For the AI, I would be more worried that it did exactly what we asked of it - People are not good at making perfect requirements on the first try. Regardless of AI or not, a team needs to be involved to have the best outcome. If you are talking about having minimal programming skills, a pile of venture capitalist loot, and a penchant for AI - I think it is a risk. Now if you have a team that knows what a good design would look like, understand how to work with an AI, and remain engaged through the process… I would be more comfortable.
Technical debt is a more complex question. Most times, technical debt grows because there is a gap between what a company can afford to do and what it wants to do - not enough time or developers. So we knowingly cut corners. With AI we can scale capacity to get what we want done. I don’t think we will see as much technical debt accruing with AI assisted coding.
However, the ease with which we can do this concerns me, and I think we will see a new unseen kind of technical debt. A new breed of developers that don’t know what to ask for and get something that won’t live up to the needs. For example, saying you need a web site to sell tickets is a lot different than needing a web site that can handle 10,000 connections a minute.
That’s my take on programming. When you look at this from a wider societal point of view, how are you thinking about it? What sort of technical debt is AI creating?
Tom
This is why, over the long term, I’m not worried about programmers who are entering the field. After we finally come to realize that AI is only useful if it’s paired with humans who have a deep understanding of the work, there will always be a place for intelligent, competent newcomers in computer science. Sure, companies can and will cut corners, and become convinced that they don’t need to hire good people, but they will either produce inferior outcomes (a practice which certainly predates AI) or realize their mistake and adjust.
My concern is that we’re just not there yet. What will happen to all of the AI-generated code being produced and distributed to consumers by people who don’t understand how it all works? If an IT department has been downsized to just a few people who need to crank out the same amount of code as a full team, they’re going to be pressured to just see if it works, and move on. That might be fine if we still have access to AI coding agents when things need to be updated, but that’s not a guarantee. At some point these corporations will either need to make money or fold, and either outcome could mean that a lot of our current capabilities will vanish.
Your question about the wider societal point of view is sobering for me, too. If these tools are calibrated to reduce friction, do the work for us, and reduce the number of decisions we need to make, that’s not exactly a technical debt. It’s more of a cognitive debt. Friction is how we learn and grow. Doing the work gives us insights and experience that we can’t otherwise obtain. How many people can resist a service that, on a surface level, makes our lives seem easier? How many will become dependent, and lose the discernment and agency that is vital to succeed in our society?
Bill
AI is being trained on what is already published, and that content is by people who are slightly above human average; that is what the code will be. This is fine for the vast majority of applications being developed out there. A lot of people talk about AI generated code as if it is in some byzantine format that no one can understand, but my experience is different. The code is not only readable, but also documented, and the AI can produce design documentation quickly and without complaint.
I don’t think AI is going away either. I do think there will be a pretty massive “market correction” where a lot of AI startups die off. But the bigger companies will continue to provide the same services as they do now, even if AI has hit its limits. I do think an era of optimization is next, where instead of struggling to produce a more advanced AI, they make the existing ones work using less power and less compute resources, which will make it easier to remain profitable. Don’t forget, there is also no vendor lock on AI generated code… meaning if Claude generates the code you can rework it later using a different vendor all together.
Cognitive debt is an interesting term. As a parallel to technical debt, it loosely becomes thinking and reasoning we are skipping now that we will have to come back to later, or suffer the consequences. Technical debt is a decision to be made and not inherently bad. I think the same applies to cognitive debt, but your words “How many people can resist a service that, on a surface level, makes our lives seem easier?” really haunt me on this one. Under the pressures a business puts on you, will good enough and quick never get revisited? If I have AI write the requirements for my next six projects, am I ever going to go back and exercise those skills? Honestly, I have my doubts. I personally try to retain the core thinking on things I do, and subject the AI to grunt work. “Here are my bullet points for a PowerPoint deck - make them pretty”, “Read my notes and produce a summary”, etc. I wonder if this is enough… focusing on the core thinking and offloading to an assistant? We started this with developers and code… for them, not coding as much anymore will definitely atrophy those skills. But is that the same as any increase in technology? We don’t forge our own nails anymore, we buy a box. Maybe the computer writing the program is the same thing.
Tom
You bring up some excellent points. I have to remind myself that long before AI turbocharged all of this, some people chose low-quality, low-effort work, and preferred to sit back while things were done for them. Just look at television for a perfect example. The critics of TV as a medium, back in the day, were often right. Others, though, aren’t content with low-effort mediocrity. They want to do the work, they insist on pushing themselves to learn and grow, and they won’t settle for junk. I’d like to live in a world where that approach is more commercially rewarding, but has it ever been that way? Perhaps AI just accelerates certain aspects of human nature that are always with us in our society, and we just have to get used to this new way of doing it. I just don’t know whether that leaves me hopeful or deflated. You’ve got the last word on this!
Bill
I get the last word after you dropped such a tasty thought there - AI as an accelerant for our own human behaviors is such an interesting view there is no way I could do it justice - I say we save that for a new thought-thread! But in the last word here - we started with the question of shipping code that AI wrote. Overall, my thoughts are that it depends on the surrounding processes… picture substituting AI for a developer, would you just ship the code anyway or is there a support process, like QA, User acceptance, etc. But AI code in itself, here in 2026 it does a pretty good job if you set it up for success. But even the best programmers don’t get a pass on testing and reviews!