AI made you faster. It didn’t make you better.
AI tools boost speed but erode quality when used to avoid thinking, leaving work hollow and reviewers frustrated.
Analysis of 1.4 million workplace conversations found only 5% of AI use improved work quality, while the rest produced output that often bypasses human review. Workers increasingly rely on AI to generate text or code, but the results frequently lack clarity, context, or meaningful insight. The shift from debating AI’s role to questioning its output reveals a growing frustration with unedited, low-value material landing in inboxes and repositories. The problem isn’t the tool itself but how it’s applied—when used to shortcut thinking, the damage becomes visible in the work others must read and trust.
Microsoft Research’s CHI 2025 study of 319 knowledge workers found those who trusted AI the most checked its output the least, while those confident in their own judgment scrutinized AI output more closely. The inverse relationship between trust in AI and scrutiny of its output highlights a systemic risk: over-reliance on AI can dull critical evaluation. Engineers report seeing AI-generated pull request descriptions that list files without explaining changes, forcing reviewers to spend extra time deciphering intent. The pattern repeats across status updates, handover documents, and risk classifications, where AI output obscures rather than clarifies.
GitClear’s analysis of 623 million lines of code changes from 2023 to 2026 shows refactoring down 70%, code duplication up 81%, and error-masking patterns like empty catch blocks rising 47%. CodeRabbit’s review of 470 pull requests found AI-generated code contained 1.7 times more issues overall and 2.74 times more security vulnerabilities. The data suggests AI accelerates production but degrades quality, with measurable increases in duplicated, risky, and poorly structured code. The trend points to a trade-off where speed comes at the cost of maintainability and security.
An MIT Media Lab 2025 study using EEG monitoring on 54 participants found those writing with an LLM showed up to 55% less brain connectivity than those writing without assistance. The same study found 83% of AI-assisted writers could not accurately recall or quote essays they had just produced. A separate MIT CHI 2026 study tracked 67 people using AI to check news over four weeks, showing short-term accuracy gains of 21% but long-term declines in unassisted accuracy of more than 15 points below baseline. The findings suggest AI can erode critical thinking skills over time, leaving users less capable when the tool is unavailable.