Margaret-Anne Storey says the industry ignores the mental models that drive software.
Traditional debt ignores mental models completely.
She proposes a framework considering three layers of system health. The model addresses technical, cognitive, and intent debt within digital products## When Code Stops Making Sense
Margaret-Anne Storey, 42, a senior engineer at a major fintech firm in London, has seen understanding fade. She watched teams build systems nobody could read a year later. Her team shipped code into roughly 900 microservices. Correctness became a moving target.
One definition fails because context changes constantly across the distributed landscape.
Team understanding erodes faster than code changes.
A developer leaves and nobody reads the original comments. The original author's intent is lost within weeks. Verification costs are rising sharply as complexity grows.
Ajey Gore argues that if coding agents make coding free, the expensive thing becomes verification. The team relies on intuition instead of deliberation daily. Kahneman's two-system model describes this split between fast and slow thinking.
Developers offload their reasoning onto tools they do not fully understand.
The stakes for a project remain high even without a crash.
Silent failures accumulate as the mental model diverges from the actual codebase. A mismatch creates risk that traditional metrics miss entirely. Teams build confidence on a foundation that is not quite solid.
Managers must account for the cost of lost context.
Verification becomes the primary bottleneck when agents handle the initial build. The industry must accept that correctness is context-dependent, not absolute.
The Cost of Lost Context
Coding agents make coding free for anyone with a keyboard. The expensive thing becomes verification now. Developers no longer write every line manually. They generate blocks of text that agents assemble in seconds.
Hundreds of engineers ship into complex microservice architectures daily. Each team defines correctness differently based on their context. One group might measure success by uptime percentages. Another prioritizes user satisfaction scores or regulatory compliance.
Legacy migration challenges surface quickly under these conditions. Teams must ensure their generated code integrates with older systems. They cannot assume the new code matches old expectations. Disagreements about what constitutes a bug arise constantly.
Someone might ship a feature that works perfectly for them. Another team finds it breaks their existing workflow immediately.
System 3 Cognition emerges as engineers uncritically rely on externally generated reasoning. They accept code without reading it thoroughly. Cognitive Offloading means trusting the agent to handle complexity.
The industry now needs new ways to validate code. Tools that check logic are insufficient. Processes that confirm intent with the original requirement become essential.
What to Measure Next
Audit mental model drift quarterly.
Developers drift from original designs as new features stack on legacy logic. These shifts happen silently over months.
Verify user goals against code today.
Requirements sit in separate files while developers implement different logic. Users expect functionality that does not exist in the build.
Refactor the invisible burdens now. Technical debt compounds when teams ignore verification costs. Cognitive debt grows as engineers rely on external reasoning.
Intent debt accumulates fastest when teams stop asking why a feature exists. The original goal fades as code evolves.
The next quarter demands a hard reset. Teams must align user goals with actual code behavior.
The regulator expects reports on verification processes by October.
