One of the most common questions about AI-assisted development is whether engineers trust AI-generated code. After reviewing hundreds of survey comments, the answer appears surprisingly simple:
Sometimes.
Trust, it turns out, is highly situational. Engineers do not trust AI equally across all types of work. Instead, trust changes dramatically depending on the workflow.
Engineers reported positive experiences when:
In these situations, AI acts as a reliable accelerator. Engineers understand what they want. AI helps them get there faster. Review remains manageable. Verification remains straightforward. Trust emerges naturally.
The opposite pattern appears when work becomes:
Engineers repeatedly described reviewing every line of generated code. Rewriting substantial portions of output. Questioning whether generated solutions were maintainable. Trust declined as uncertainty increased. Not because engineers disliked AI. Because the cost of being wrong increased.
One of the most interesting findings from the DevEx survey is that trust appears to be influenced by more than correctness. Understanding matters. Engineers trust code they understand. They distrust code they cannot explain. This is why many comments focused on large AI-generated changes becoming difficult to review. Even when generated code appears correct, confidence drops when understanding drops. Trust and comprehension are deeply connected.
Trust directly affects delivery.
When trust is high:
When trust is low:
This means trust is not simply a psychological factor. It is an operational factor. It influences how much of AI’s productivity actually reaches production.
The survey suggests that trust is not really a property of the model. It is a property of the workflow. The same engineer may trust AI completely in one situation and distrust it entirely in another.
The difference is usually:
This helps explain why some teams experience significant benefits from AI while others struggle. They’re not necessarily using different tools. They’re operating in different trust environments.
The question is not whether engineers trust AI. The question is: Under what conditions do engineers trust AI? That distinction matters because the future of AI-assisted development will depend less on universal trust and more on understanding the situations where trust can be earned, maintained, and justified. We can help.
