A Tester With Seven Years of Experience Examines the Software Quality Crisis Thesis

A black screen after a Windows update, a rollback, relief. That is how an article by a Cloud.ru QA engineer begins; he has been testing software for seven years. His conclusion is cautious: one painful failure, even one experienced personally, says nothing about the state of the whole industry. The title, however, asks the question bluntly: "Global software quality crisis: is it really that bad?" The published announcement does not answer it; the full text is on Habr. We will not retell what we have not seen. Instead, we will check the question against data that is already publicly available.

What the numbers say

The "software got worse" thesis has strong arguments behind it. In July 2024, a faulty CrowdStrike update took down about 8.5 million Windows devices, according to Microsoft. Airlines, banks and hospitals hit the same error at the same time. In 2012, Knight Capital lost about $440 million in roughly 45 minutes because of incorrectly deployed code. And the CISQ consortium estimated total US losses from poor software quality at $2.41 trillion for 2022. That figure should be treated as a model, not an accounting report, but the order of magnitude is telling.

Skeptics have their own counterarguments. We see more failures because there is incomparably more software now, and any incident instantly lands on social media. A "before and after" comparison is hard to build: there is no single metric of software quality.

Why personal experience misleads

The tester's position is methodologically sound. A failure after an update sticks in the memory, while thousands of updates that went smoothly go unnoticed. This is the classic availability bias: a vivid case is perceived as a trend. QA work consists precisely of replacing impressions with measurements.

There is plenty to measure:

• the share of releases that required a rollback or hotfix;
• time to recover from an incident (MTTR);
• the failure rate of changes (change failure rate);
• the number of critical defects that reached production;
• the hourly cost of downtime for a specific business.

On these measures, the industry is diverging rather than moving in one direction. Teams with mature CI/CD and automated testing ship releases more often and more reliably. The rest gain speed at the expense of reliability.

Where AI comes in

One factor of recent years that the announcement does not mention is code generation. The DORA 2024 report showed an unwelcome correlation: a 25% increase in the use of AI tools came with a drop in delivery stability of about 7.2% and a 1.5% drop in throughput. Correlation does not prove causation, but the signal matters: code gets written faster, while verification does not speed up at the same pace. Review and tests become the bottleneck, and this is where a tester's work is valued more than before.

How to use this question in practice

Do not make decisions based on headlines about a "crisis." If you are responsible for a product, calculate your own metrics: change failure rate and MTTR over the last two quarters will tell you more than any industry estimate. If you are rolling out AI coding assistants, plan for a rise in review and automated-testing load at the same time as the rise in development speed. And for updates to critical infrastructure, keep staged rollouts and a proven rollback plan. That plan is what got the article's author his working computer back.