A recent DEV Community post by @xulingfeng (https://dev.to/xulingfeng/we-keep-hearing-about-ai-productivity-the-qa-group-chat-told-a-different-story-2nma) follows up on an earlier article about the author's company's AI testing practices, using reactions from a QA community group chat to question how AI productivity claims play out for the people doing the testing.
An anonymous poll and its limits
In a roughly 500-member QA group, an anonymous poll asked whether members were currently employed. Of 76 respondents, 60 (78.9%) said yes and 16 (21.1%) said no. The author is careful to note this is not an industry unemployment survey: the poll was anonymous, the reasons for those 16 responses are unknown, and only a fraction of the group voted. Still, the author mentions a former colleague in software testing who has been out of work for six months, and quotes group comments about layoffs, age, and whether the field still has a future.
Can AI compare an app screen to its design?
A member asked whether AI test automation could compare an actual app screen against its UI design, checking sizes, colors, and positions. Suggestions included OCR and multimodal models, alongside concerns about maintenance when the UI changes. The author's answer: it is technically possible, but the model needs enough context and detailed instructions, and top-tier multimodal models can consume a lot of tokens. The author's pointed question was whether management is paying for those tokens, arguing that if a company wants AI-driven productivity, AI usage belongs in the company's budget rather than on employees' own credits.
AI writes the code, AI tests it
Another thread described an almost fully automated pipeline: generate a demo and PRD, generate architecture and code, generate tests, ship it, and "a pile of bugs goes live." The author argues a working demo is not a validated product, and that generated code and generated tests do not automatically share the same understanding of the system.
A concrete example from the author's own testing work: in a knowledge-base project, AI found 36 potential defects and manual testing found 11; after review, 30 were confirmed valid, 21 of them from AI. Some reported issues were not implementation bugs at all, but cases where product and development had agreed on changes without updating the PRD. The AI followed the document; the people followed an agreement that existed only in conversation.
Why the rush, and who benefits
The post also covers group discussion of companies copying competitors' AI adoption, pressure from large vendors' full-stack AI messaging, and a cynical view that token sellers have an interest in promoting the productivity narrative. The author does not go that far, noting personal use of these tools in testing and code analysis, but points to commercial incentives that can reinforce one another even without a worked-out business case. One member suggested AI may be a trigger that makes an existing hiring imbalance more visible rather than its original cause.
The closing line the author repeats from the chat: embrace AI and get fired later; don't, and get fired now. The post does not resolve that tension, but it makes the case that productivity claims should be examined alongside who pays for the tools and who absorbs the risk.