AI project prep has become a common first step for clients working with a design studio. There’s a video making the rounds that stopped me mid-scroll, not because it was surprising, but because it was exactly right. It’s the best illustration I’ve seen of why AI project prep — however capable the tool — still needs the right context and a human in the loop.
Creator Nathan Doan posted a reel where he asks an AI to cut a hoagie in half. Simple enough. Except the AI cuts it lengthwise. He says “no, cut it the other way”… and it just flips the sandwich and cuts it lengthwise again. Same action, different orientation, same wrong result. It takes several rounds of increasingly specific prompting to get what anyone standing in a deli would have understood in two seconds. 
It’s funny. It’s also one of the clearest illustrations we’ve seen of how AI actually works — and why the gap between what you ask for and what you get can be wider than it looks. The tool was capable. The intention was clear. But without the right context, the right framing, the right specificity… it just kept confidently doing the wrong thing.
Watch the original reel here: @nathandoan on Instagram
We’re seeing a version of this play out in how clients come to us now. People are using AI to scope projects, build briefs, research timelines, and generate budget estimates — and a lot of that prep is genuinely useful. The clients who arrive having thought through their project in advance make for better conversations and faster starts. That hasn’t changed.
What we’ve learned, after working through enough of these engagements, is that AI-assisted prep is only as strong as the instructions behind it. When someone prompts a tool without full knowledge of what a studio actually needs — or how a real project actually moves — they end up with a plan that looks complete but was built on assumptions that were never examined. Confident. Detailed. And sometimes, pointing the wrong direction entirely.
Here’s what that looks like in practice, and how to make your AI project prep work harder when you bring a real team in.
Context is the variable AI can’t account for
A language model will generate a reasonable project scope based on what you describe to it. What it can’t do is factor in where your business actually is, what your market is doing right now, or industries over years of real work. That layer of institutional knowledge is the gap — and it’s a significant one.
We’ve worked through engagements where a client arrived with a technically sound plan — well-structured, logical, thorough — that needed meaningful repositioning once we understood the full picture. The AI built something sensible for the problem as described. The problem, as described, wasn’t quite the real one. That’s not a failure of the tool. It’s the natural result of scoping before a real conversation has happened, with instructions that couldn’t include what the client hadn’t yet said out loud.
The brief is a starting point. Discovery is where it becomes a direction.
The most useful thing you can bring to a first meeting isn’t a finished plan. It’s a clear articulation of what changes in your business if this project actually works.
AI pricing reflects a generic project — yours isn’t
This comes up consistently. A client runs their project description through an AI tool, receives a cost estimate, and arrives with that number as their anchor. The figure is usually derived from industry averages and general market assumptions — a composite of similar projects, not an assessment of the actual one in front of us.
What that number doesn’t account for: the specificity of your brand, the revision cycles built into real creative collaboration, the strategy layer that determines whether the execution lands, and the accountability that comes with a studio putting its reputation behind the work. Prompting an AI to price a brand project is a little like asking it to cut the sandwich — it’ll give you an answer that technically addresses the question, but it’s working without the context that changes everything.
Bring the estimate to the table. It tells us how you’ve been thinking about the project and gives us a useful place to start the conversation. Just know that what we build from there is based on your actual project — not a version of one.
When the plan shifts mid-project, say so
One of the more nuanced dynamics we navigate is when a project quietly changes direction. A client takes notes from a strategy session, runs them back through an AI tool, and returns with what reads like a new recommendation — sometimes a restructured scope entirely — without flagging that the source has changed.
It happens because AI is available at 11pm when an idea is still forming and studios aren’t. That’s a real and understandable reality. The challenge is when that AI-generated pivot gets absorbed into the project without being examined — because the model generating it didn’t have the full context of what the studio already scoped, already decided, already built. The plan gets re-prompted, the result comes back confident and detailed, and suddenly everyone is working from different foundations without knowing it.
The more productive version of that same instinct: bring it back to the room. “I ran our strategy session notes through an AI and it flagged a different approach to the architecture — can we look at this together?” That’s a conversation worth having. It keeps the project moving from shared ground instead of parallel ones.
A studio relationship moves fastest when communication is direct. If something shifted, if you got new information, if you’re second-guessing a direction — say it. That’s exactly what the relationship is for.
Use AI to sharpen your questions, not close them
The best use of AI in the pre-project phase isn’t generating answers — it’s generating better questions. Clients who use it that way walk into discovery calls knowing what they don’t know yet, which makes for a sharper, faster conversation on both sides.
We’ve had clients use AI to build vocabulary around technical concepts they weren’t familiar with so they could actually follow and contribute to architecture discussions in real time. That’s the tool working well. The project moved faster because we weren’t spending the first few sessions getting everyone oriented from scratch.
Where it creates friction is when AI-generated conclusions get treated as decisions. Technology stack choices, timeline commitments, feature prioritization — these carry downstream implications that require someone with full project context to weigh in on. That’s the work a studio does, and it needs to happen before those calls are locked.
How to make your AI project prep work harder
- Use AI to clarify your thinking — then bring that thinking to a real conversation
- Hold AI-generated estimates as a reference point, not a ceiling
- Flag when your brief or scope was AI-assisted — it helps us calibrate faster
- Bring AI-generated pivots back to the studio before acting on them
- Use AI to prepare questions, not finalize answers
- Stay open to the discovery process — that’s where the real scope lives
The projects that go well start the same way
After enough engagements, the pattern becomes clear. The projects that move well and deliver strong outcomes almost always start with a client who came prepared but stayed curious. They had a point of view and they were willing to have it challenged. They used AI to get oriented — and then they walked into the room ready to let the work evolve beyond what a prompt could produce.
Think back to the sandwich. The person knew what they wanted. The AI was capable of helping. The missing piece was the specificity, the context, the back-and-forth that gets you to the actual result. That’s not a knock on the tool — it’s just an honest picture of what it takes to get from a good idea to something built right.
We’re the next layer in that process. Bring your AI research, your estimates, your half-formed thinking. We know what to do with all of it — and we know which way to cut.
