Jonathan Haaswritingnowusesabout
emailgithubx
Jonathan Haaswritingnowusesabout

Testing at Light Speed: How QA Adapts to AI Velocity

June 28, 2025·3 min read

AI-generated code produces different bugs than human-written code. QA built for syntax checking is testing for the wrong failures.

#qa#testing#velocity#ai#quality-assurance

"How can we possibly test features that are built in hours?"

A QA lead asked me this after their dev team started using AI pair programming. They went from quarterly releases to daily deployments. Their testing process -- built around batch review cycles with weeks of buffer -- collapsed.

The speed problem is real but secondary. The deeper issue: AI-generated code produces fundamentally different bugs than human-written code, and most QA processes are testing for the wrong failures.

AI Makes Different Bugs

AI produces perfect syntax, consistent patterns, correct error handling, and thorough documentation. It rarely makes typos or forgets a semicolon.

What it gets wrong: business logic assumptions (misunderstanding requirements), context confusion (applying patterns from the wrong domain), subtle security vulnerabilities that are functionally correct but exploitable, performance characteristics that only surface at scale, and edge cases specific to your domain that aren't in the training data.

The code looks perfect but may contain deep logical flaws. Traditional syntax-focused testing catches none of this. Business logic validation catches all of it.

Test Where AI Fails

The strategy is simple: stop testing what AI is good at. Focus human effort on what AI gets wrong.

Risk-based prioritization. Payment processing, security boundaries, data migration, and third-party integrations get thorough human review. UI layout and content updates get automated checks only.

Business logic verification. The most dangerous AI bugs are "plausible but wrong" -- code that implements the incorrect business rule while looking completely correct. QA must verify that behavior matches requirements, not that code compiles.

Domain-specific edge cases. AI builds currency converters that assume 2 decimal places for all currencies. It builds auth flows that store tokens in localStorage. It builds data processors that work with 100 records but fail at 1 million. These are the bugs worth hunting.

Shift Testing to Production

Instead of catching everything before deployment, invest in catching things fast after deployment.

Feature flags for gradual rollouts. Real user monitoring. A/B testing for UX validation. Performance monitoring under actual load. The target metric is minutes-to-detection, not months-of-prevention.

This doesn't mean shipping untested code. It means accepting that production is a testing environment too, and building the infrastructure to make that safe: instant rollback, circuit breakers, real-time alerting.

QA Becomes Infrastructure

The biggest shift: QA stops executing test cases and starts building testing infrastructure.

Manual regression testing, requirements translation, bug reproduction, test case documentation -- these activities shrink or disappear. What grows: continuous testing pipelines, real user behavior monitoring, support ticket pattern analysis, and coaching developers on quality practices for AI-generated code.

The QA role transforms from Test Case Executor to Product Quality Advocate. The focus moves from defect detection to outcome optimization.

Old thinking: "Our job is to find bugs before customers do."

New thinking: "Our job is to ensure customers have great experiences."

The teams that make this shift ship better products faster. The teams that cling to batch-processing QA become the bottleneck their organizations route around.

share

Continue reading

Two Minds in the Machine: Shared Context Is the Only Thing That Matters

I added Gemini to a codebase that already had Claude embedded. The useful discovery was about shared context files, not model capabilities.

AI Detection Hysteria: When Human Creativity Gets Mislabeled

A photographer friend posted a sunset photo after three hours of waiting for the perfect light. Within minutes: 'Obvious Midjourney.' 'Nice prompt, bro.'

The CLI Renaissance: How AI is Driving the Command Line Revolution

AI coding assistants output shell commands, not GUI instructions. That single fact is reversing a decade of developer tooling trends.

emailgithubx