Why AI Feels Broken Early On, and How to Train Past the Dip
A designer argues that early disappointment with AI is a training phase, not proof the tools fail, and explains how to climb out of it.
Pierre Silva1 min read
Treating the Bad First Result as Training
This essay takes aim at two opposite habits: dismissing AI after one poor output, and assuming it can handle any task with a single prompt. The author argues that both views expect a magic button. The real work is building your own ways of working with the tools, shaping inputs, briefs and examples until the output matches your standards. The author describes a design workflow covering research, briefing, synthesis and option generation, and admits that a half day of effort can still produce results that miss the mark. That moment is where most people give up.
The piece leans on the idea of a trust ladder, borrowed from a conversation about agent tooling, to explain why early errors shrink confidence and why patient iteration rebuilds it. The helpful framing is to ask what to change in the instructions or the inputs, rather than whether the tool works at all. For brand builders, product designers and strategists who are testing AI in their process, this is a grounded read on expectations, training time and what trust in a workflow looks like in practice. It is worth a few minutes before you judge a tool on a single attempt.
Source: muz.li · Originally published
