The First Real Feedback

When someone else uses the thing you built

There’s a version of building where you’re the only one evaluating the output. You know what the tool is supposed to do, you know where the rough edges are, and you know which parts to avoid when you’re showing it to someone. That version ended this week.

Mid-term presentations meant putting both tools in front of our supervisors and the people who would actually be coordinating how they get used. Not a structured test, not a controlled demo. Just: here’s what we built, what do you think.

Feedback from people who know the workflow the tool is supposed to fit into hits differently than feedback from someone seeing it cold. They’re not just evaluating what it does — they’re evaluating whether it fits where it needs to go.

What changed on the image side

A lot of this week was polish — but polish with a clearer direction than before. The feedback made it obvious that a tool built by the people using it will always have blind spots. Features that need explanation stop feeling like they need explanation when you’re the one who built them. Interactions that are confusing become invisible when you already know what they do.

So the focus was on making the tool make sense to someone who wasn’t there for the build. The brand kit got restructured into a tabbed layout so everything has a logical home and isn’t competing for space. Asset management now has proper categories, search, and the ability to organise things the way a real brand workflow would — not the way it happened to get built. Visual identity styles are editable inline now, which removes a step that made sense architecturally but didn’t make sense in practice.

A handful of bugs that had been tolerable internally got prioritised because they wouldn’t be tolerable for someone using the tool for the first time. Multi-select behaviour, font loading delays, icon rendering issues — small things that add up to a tool that feels unfinished when you hit them without context.

The user guide is being built directly into the system for the same reason. Prompting advice, system guidance, the things that currently live in our heads. The goal is that someone can pick it up without needing us in the room.

Most of the feedback from the presentation was less about what was broken and more about what the tool could become — use cases, features worth adding, directions worth exploring. That’s a good place to be at the halfway mark.

What changed on the web side

Getting feedback on something you build is a strange experience. You know it is useful and you know it is how the product gets better. Even so, hearing that something feels generic after spending days on it is never easy.

That was exactly where we found ourselves with the first version of the tool. It worked well from a technical perspective. The sites were clean, responsive, and well structured. The problem was that they all had the same feel. I was giving Claude prompts like corporate and trustworthy and expecting it to understand what made each business unique. Without enough context it defaulted to the safest answer.

We changed the process instead of expecting better results from the same prompts. The tool now generates real design components before a page is built, including typography, buttons, cards, and spacing. We also added a messaging step so the client’s positioning is defined before any copy is written. Competitor analysis gives the AI more context about what already exists and how the new site should stand apart.

One of the most useful additions came directly from feedback. Every brief is now scored across messaging, branding, goals, design, pages, forms, and analytics before it can be exported. Missing information is highlighted so everyone knows exactly what still needs work. We also added a style preview that renders every design direction using the client’s own colours and fonts, making it much easier to choose the right direction before anything is generated.

The biggest lesson has been that feedback usually exposes assumptions, not failures. Every time something felt off it pointed to information the AI should have been given instead of being expected to guess. Fixing those gaps has improved the tool far more than rewriting prompts ever did.

Where both projects go from here

Feedback week and integration week are arriving at the same time, which is convenient. Now that both tools have been through a real presentation and come out with a clearer sense of direction, the next question is how they actually fit into the way Goose Digital works.

That means drafting workflows. Not building them yet — mapping them. Where does the tool enter the process, where does a human take over, what needs to exist in between. The brand intelligence schema that both tools share is the connective tissue, and getting that integration right is what determines whether these stay as standalone tools or become something that’s actually embedded in the day to day.

The second half of the co-op has a shape now. Polish, integrate, test with real clients, and figure out what production actually looks like. The feedback gave both projects a clearer sense of what they’re building toward. That’s probably the most useful thing a mid-term can do.