---
id: "claim-latent-ai-errors"
type: "claim"
source_timestamps: ["¶10", "¶11"]
tags: ["risk-management", "technical-debt"]
related: ["claim-code-vs-engineering", "action-escalation-rules", "question-insurance-pricing"]
confidence: "high"
testable: true
speakers: ["Chengwei Liu", "Balázs Kovács"]
sources: ["futures"]
sourceVaultSlug: "hbr-seg-futures"
originDay: 2
articleStem: "hbr-cl-84-big-tech-capability-crisis"
sourceUrl: "https://hbr.org/2026/06/big-techs-looming-capability-crisis"
sourceTitle: "Big Tech’s Looming Capability Crisis"
---
# AI Coding Errors Have Highly Latent Impact Timelines

## Claim: AI Coding Errors Have Highly Latent Impact Timelines

**Confidence: high · Testable: yes**

Unlike radiology — where an error harms a **named patient quickly**, with a clear chain of liability — AI errors in software coding **surface years later**. Bad code can look perfectly functional at launch; defects typically become apparent only when the system needs to be **modified, integrated, secured, or scaled** — long after the original prompt is forgotten and the author has left the company.

This latency is why the authors argue software lacks medicine's disciplining feedback loop, and why they propose [[action-escalation-rules|escalation rules]] to force accountability. It also feeds the [[question-insurance-pricing|open question]] of how insurers will price these delayed defects.

> Enrichment: Plausible and partially supported, but this is an **analogy rather than a measured comparative study** — software failures often surface later during integration, maintenance, scaling, or security changes.