BuildingwithAI:WhattheWorkflowActuallyLooksLike
Everyone's talking about AI and design. Here's what the actual day-to-day workflow looks like when you're using these tools to build a real product, not just a demo.
There's a lot of talk about AI and design right now, and most of it is either breathless hype or defensive dismissal. Neither is useful if you're actually trying to ship something.
We build with these tools every day. We used them to build Bento Japanese, our own product, from scratch. We use them on client work too. So instead of another opinion piece about whether AI is "coming for design jobs," here's what the workflow honestly looks like, including the parts that don't work as well as the demos suggest.
Where AI actually helps
The biggest shift isn't that AI does the design or the coding for you. It's that it collapses the distance between thinking and building. Historically, the gap between "I have an idea for how this screen should work" and "I can see it on screen" involved a lot of mechanical steps. Setting up components, writing boilerplate, wiring up basic interactions. None of that work required much judgment, but it all took time.
AI tools are very good at that part. You describe what you want, and you get a rough version fast enough to react to it instead of just imagining it. That changes how you think. You can test five directions in the time it used to take to build one, which means more of your time goes into actual decisions instead of production.
Where it falls apart
The part nobody puts in the demo video is how often the first output is confidently wrong. Not broken, which would be easy to spot. Wrong in a way that looks plausible. A component that almost matches your design system but uses a slightly different spacing scale. A user flow that technically works but skips an edge case that matters for your specific users. Code that runs without errors but handles a case incorrectly in a way you only notice three steps later.
This is the part that requires actual experience to catch. If you already know what the right spacing scale is, what the edge case should be, what correct behavior looks like, you spot the problem in seconds and move on. If you don't have that judgment yet, you don't know what you're looking at. You either accept something subtly wrong, or you spend hours debugging something that an experienced eye would have caught on sight.
This is why we think the framing of "AI replacing designers and developers" misses what's actually happening. It's making the gap between people who know what they're doing and people who don't much bigger, not smaller.
What the actual workflow looks like, end to end
For Bento Japanese, and increasingly for client projects, the loop looks roughly like this. We start with the thinking that has nothing to do with any tool: who is this for, what does it need to do, what does success look like. That part doesn't get faster with AI and shouldn't.
Once the direction is clear, we move into rapid prototyping. This is where tools like Figma and AI-assisted design tools earn their keep, turning a concept into something visual fast enough to react to and refine before committing real development time to it.
From there, building happens largely through Cursor. We describe the feature or component, get a working version, then review it the way you'd review a junior developer's pull request. Not rubber-stamping it. Actually checking the logic, checking it against the design intent, catching the parts that look right but aren't quite. That review step is the part that takes real skill, and it's the part that's easy to skip if you don't know what you're looking for.
The cycle repeats. Build, review, adjust, ship a piece, move to the next. The speed comes from compressing each cycle, not from skipping the judgment in between.
The honest takeaway
If you already know what good design and good code look like, these tools make you faster, more capable, and frankly more dangerous in a good way. You can do in a week what used to take a month, and you learn faster because you're seeing more attempts.
If you don't have that judgment yet, the same tools can quietly waste your time. You'll produce something that looks finished and isn't, and you won't know which parts to trust. That's not a reason to avoid the tools. It's a reason to be honest about where you are, and to bring in experienced judgment for the parts where being wrong is expensive.
We don't think that's a controversial take. It's just not the version of the story that gets clicks.
Book a free 30-minute consultation → if you're trying to figure out where AI tools can genuinely speed up your project, and where you still need experienced hands on it.
