As AI and agents are adopted to accelerate development, cognitive load and cognitive debt are likely to become bigger threats to developers than technical debt
The term technical debt is often used to refer to the accumulation of design or implementation choices that later make the software harder …
Margaret-Anne Storey
Context & Ripple Effects
Earlier coverage warned that generative AI's lower barrier to code creation could bring more complexity, debt and confusion, while accumulated quick fixes and legacy code show how unresolved software choices can become operational and security liabilities.
This article shifts the focus from code quality alone to developers' ability to understand, verify and maintain faster-moving systems. That concern sits alongside later reporting of longer hours among workers offloading tasks to AI, complicating a simple productivity narrative.
First-order effects
Teams adopting coding agents must devote more attention to reviewing AI-produced changes, preserving system context and documenting decisions—not merely measuring output speed.
Developers face a higher risk that knowledge of why a system works becomes fragmented across prompts, agent runs and rapidly generated code, making maintenance harder.
Second-order effects
Engineering leaders are pressured to add stronger review, testing and knowledge-sharing practices, which can offset some of the apparent time savings from agent-assisted development.
Tool vendors and platform teams gain an incentive to make agent actions, provenance and handoffs easier to inspect, as existing concerns about AI-created complexity and confusion become a workflow problem.
Third-order effects
If agent-assisted output continues to outpace teams' capacity to comprehend it, software delivery may increasingly be constrained by organizational understanding rather than code-generation capacity.
The durable differentiator could shift toward governance and system knowledge: firms that can retain context, accountability and maintainability may capture more value from development agents than those optimizing for volume alone.
The trend: AI-assisted software development is moving the bottleneck from producing code toward sustaining human understanding, oversight and accountability around it.
“Cognitive debt is likely a much bigger threat than technical debt....” https://margaretstorey.com/... “As generative and agentic AI reshape how software is built, understanding and managing cognitive debt may be one of the most important challenges our field faces.” [image]
Short musings on “cognitive debt” - I'm seeing this in my own work, where excessive unreviewed AI-generated code leads me to lose a firm mental model of what I've built, which then makes it harder to confidently make future decisions simonwillison.net/2026/Feb/15/ ...
A complication of AI-assisted coding: developers (even if they're being careful, checking everything, etc.) are moving too fast to develop mental maps of their projects and remember their design and architectural choices, making it harder to explain those choices or make changes …
Cognitive Debt. — “Even if AI agents produce code that could be easy to understand, the humans involved may have simply lost the plot and may not understand what the program is supposed to do, how their intentions were implemented, or how to possibly change it.” — margaretsto…
Let that sink in. Even one of the loud and vocal AI proponents such as @simon admits — “I no longer have a firm mental model of what they can do and how they work, which means each additional feature becomes harder to reason about, eventually leading me to lose the ability to …
“managing LLM agents has a lot of similarities to managing junior developers” — No, they are different: LLM agents don't make coffee but on the other hand you can shout at them without being attacked as a toxic manager. — [Otherwise, interesting article.] — https://martinfo…
This whole thing is a good read. I've often referred to “cognitive overhead” when working in engineering organizations pre AI to explain why velocity can grind to a halt, but “cognitive debt” resonates more when it is being deliberately created by AI as opposed to developing ove…