A widely discussed piece from design researcher Vitaly Friedman pushes back on the assumption that users are hungry for more AI features in the products they use every day. The argument, drawn from UX research and industry data, is that low adoption and weak retention of AI features often trace back to how those features are built and delivered, not to any inherent user hostility toward AI.
The Problem With Bolt-On AI
Many AI features are added as separate tools sitting outside a person’s normal workflow. Instead of removing friction, they add another system to switch between. The analysis points out that verifying AI-generated output carries real costs: skimming the response, spotting key points, checking them one by one, confirming the reasoning, and often correcting and regenerating the result. That review cycle can eat up any time saved by generating the content in the first place.
The piece also cites productivity research suggesting AI use can intensify work rather than reduce it, and notes that AI tends to amplify existing organizational problems, such as poor data quality or unclear decision-making, rather than fixing them. When those problems surface through an AI tool, users are left to sort out the mess themselves.
What Actually Works
According to the analysis, people are not comparing AI to human unreliability. They are comparing one feature to another and choosing whichever works consistently. The consistent theme across sources is that AI succeeds when it is:
- Deeply integrated into an existing workflow rather than a separate, bolted-on tool
- Focused on automating mundane, repetitive tasks rather than replacing an entire process
- Fast, accessible, reliable, and predictable every time it is used
- Aligned with the mental models and habits people have already built for their work
Why This Matters For WordPress Sites And Hosting Teams
For WordPress site owners and hosting providers rolling out AI-powered features, such as content generation plugins, AI-driven support widgets, or automated site management tools, the takeaway is straightforward. Adding an AI badge to a feature does not create value on its own. Features that require users to leave their normal editing or admin workflow, verify output line by line, or manage new agent-driven processes are likely to see the same low adoption patterns described in the source research.
The more durable approach is to target genuinely tedious tasks, such as image alt text generation, log summarization, or routine moderation, and make sure the automation fits inside tools people already use rather than asking them to adopt an entirely new interface.