Agentic Development is always Tech Debt
But whether this is good or bad hangs on WHY it's being incurred.
Tech debt gets a bad rap. For many engineers, the term is synonymous with bad code — poorly architected, deeply coupled, difficult to understand software.
But it hasn’t always been this way. When Cunningham first coined the phrase, he framed it in the financial sense. Shipping code that you know isn’t quite right now, is like taking on financial debt. You get a short-term boost but pay “interest” later with the extra effort to change or extend the code.
It’s a way of spending tomorrow’s engineering resources today.
If you think of tech debt as a temporal loan, then it is neither good nor bad. Like any tool or instrument, its value is derived exclusively from how it is used.
Used wisely — to get a new feature to market and validated earlier than otherwise possible — taking on tech debt can be a smart choice. You’ll pay a higher total cost than doing it right the first time, but you’ll get earlier feedback or first move advantage for the trade-off.
Used poorly — taking shortcuts just to save money in the short term — taking on tech debt isn’t much different to taking out an unsecured loan for your daily living expenses.
Agentic development gives us a very similar trade-off to the tech-debt-as-a-temporal-loan view. It gives us a powerful new tool to fast-track code generation. It allows an experienced engineer to significantly leverage their knowledge and skills, across a range of your languages platforms where they may lack specific syntactic knowledge.
But agentic development also introduces its own problems. The speed and volume of code generation means that human review struggles to keep up. LLM generated code is more verbose and cognitively complex than human generated code.
The risk is real — Anthropic had to patch 133 bugs in this week’s release alone.
None of this by itself is problematic however.
If used deliberately, LLM-generated tech debt, like human-generated tech debt, can be a sound strategic play — exploiting short-term opportunities for greater long-term cost.
But there are two important differences to keep in mind. The first is that LLM-generated tech debt, is just as costly as human-generated tech debt - but generated 100 times faster.
The second is that human-generated tech debt can be deliberate but LLM-generated debt can’t. LLM’s lack intentionality. They can’t make judgements and decide if one implementation is more prudent that the other. They just do — generating the most statistically likely next tokens for the given context.