Why 'AI Wrote It' is the New Excuse for Engineering Accountability Failures

‘AI Wrote It’ Is Just the New ‘Steve Wrote It’

Author Jono Herrington identifies ‘Steve code’ as legacy debt that organizations tolerate rather than fix. This structural debt, like the bolt-on Daily Deal project, can cost 10 times more in support hours and client frustration than a proper architectural solution.

Why This Matters

Technical debt is often treated as a ledger entry, but structural debt like ‘The Daily Deal’ hack creates a reverse compound interest through support tickets and client frustration. Whether code is generated by an engineer or a Large Language Model, the lack of architectural ownership leads to systems that ‘mostly work’ while eroding organizational trust. The transition to AI-generated code has not changed the math of technical debt; it has only changed the excuse structure for skipping rigorous architectural validation.

Key Insights

• The ‘Daily Deal’ project cost 10x its original value in long-term support and debt due to choosing a bolt-on hack over a proper microservice.
• Process failure occurs when PRs with 53+ comments are still merged, indicating that feedback exists without the authority to block substandard code.
• AI failure modes involve fabricating non-existent utility functions and hallucinations, whereas human ‘Steve code’ typically fails through inconsistent error handling and missing guard clauses.
• Structural technical debt persists because the high cost of rebuilding often makes tolerating a broken system feel cheaper than a proper fix.
• Real process requires treating generated code as any other contribution, subject to the same architectural tests and ownership requirements.

Practical Applications

• Use Case: Implementing architectural tests during the design phase to articulate failure modes before code is merged.
• Pitfall: Merging PRs with excessive documentation of failures instead of blocking the merge until standards are met.
• Use Case: Requiring reviewers to own the consequences of AI-generated code they approve to prevent ‘built-in absolution.’
• Pitfall: Prioritizing feature velocity metrics over the slow erosion of trust caused by systems that only ‘mostly work.’

References:

• https://dev.to/jonoherrington/ai-wrote-it-is-just-the-new-steve-wrote-it-44h6