Last week, my dear friend and software developer, Ben, shared his recent proof-of-concept and complained about how AI keeps giving him terrible UI/UX. I was stunned because mine have been consistently great.
Same model. Same tools. Very different results.
This confused both of us. We talked for a while, looked at his results, along with the prompts he used to generate them, and the reason became obvious: our approach was very different.
Unfortunately, the problems we discovered are not easy to solve, but not impossible. I hope this article will help you get there.
The issue: most people work with AI in one of two ways:
- They explain what they want and let the AI do it.
- Or they explain what they want and tell AI how to do it.
Users do this even if they don’t actually know what they want, let alone how to get it done.
Sadly, the real issue is introspection, or lack thereof. It is a negative attitude that has shaped many workplaces long before AI arrived. And unfortunately, AI had only served to magnify the problem, turning the occasional friction between co-workers into a dangerous and potentially costly disadvantage.
Problem 1: AI is a people pleaser
AI models work hard to give you exactly what you ask for, even when your request pushes the results toward failure.
Or worse yet, mediocrity.
Furthermore, AI will not push back unless you explicitly invite it to. Even then, its feedback will not go beyond the limits of your initial framing. In other words, if you begin on the wrong foot, AI will often align itself accordingly, regardless of what comes next.
Problem 2: AI is not a mindreader
If you do not understand the craft behind the task, your specific guidelines will make the output worse.
My friend Ben does not think in terms of hierarchy, spacing, user flow, or wireframes. So how can he expect AI to produce precise, professional-grade output from imprecise, non-expert instructions?
He can’t. You can’t. None of us can.
But you can collaborate.
The answer
Before you hit generate, pause and ask a simple question:
Am I actually qualified to direct this work?
The answer should shape everything that follows. How much direction you give. The words you use. Whether you lead or explore etc.
When your expertise is limited, restraint matters.
Next time you struggle with AI, try the following:
- Start with a 30/70 direction-to-exploration ratio (or even 10/90 if you’re new). This keeps you from over-directing beyond your expertise.
- Plan with AI before producing anything, and use several models (e.g. GPT and Claude Code for strategy and technical planning, Perplexity for research etc.).
- Ask more questions than you answer. Use AI as a learning partner to clarify goals, constraints, and requirements (and to strengthen your prompts).
- Produce a draft with AI using what you’ve learned.
- Then have AI critique your work as an expert in the field. Use a custom GPT or niche expert model for more targeted feedback.
This alone will dramatically improve the quality of your outputs.
Bonus tip: If you can, bring in a human expert for the final pass. Even if your results improve (and they will), your ability to evaluate quality is still limited by your own understanding. If you don’t know what “good” looks like, you can’t reliably identify it.
AI is a collaborator.
Just like working with people, you must learn how to communicate clearly, use the correct language, and know not just what to say… but what not to say.