We failed, So you can learn

AI... AI... AI... Ever since ChatGPT launched, everything online seems to mention AI at least once a day. Clients come to us constantly asking for AI solutions. We use it every hour at work, even before AI was actually good.

But we work in tech, so our work involves moving digital bits. AI, trained on trillions of tokens, fits naturally into what we do. Most traditional businesses, however, might not have a meaningful use case for AI yet.

Things look great on the first try - but not when you attempt to put them into production.

A long-time client runs an ecommerce shop and uses industrial embroidery machines. They take many custom orders and employ sketch artists to draw line diagrams from customer images. They wanted to quickly generate outlines from customer-submitted images. These outlines are then digitized and fed into embroidery machines.

Most image AI models can produce beautiful outlines. But just like text models that never hit 100% accuracy, image models often get you 90% close, not perfect.

In this case, the problem was far more specific.

Embroidery machines can only stitch in a continuous line. That means the outline must be one single, unbroken path. No gaps. No breaks. No tiny wiggles.

This requirement never surfaced during discovery. The client liked initial visual results, but almost every outline had disconnected segments, not one continuous line.

No amount of prompt engineering could force the model to reliably generate a single-stroke outline. It's simply not what these models are designed for.

The client runs an active, profitable embroidery business, and delays would create even more pressure on an already strained workflow.

So we improved our consultation process. Now we go deeper on expectation-versus-reality conversations around AI. We explain what AI can do today and where limits still exist.

AI is improving at an exponential rate, and these challenges may be resolved someday. But we also have to remember: fast software has existed for years, and many businesses are still far from using tech's full potential.