A build shows what a report cannot.
Focus and relevance are the new hard part. Let me back up.
Last year, the frustrating part of a research recommendation was getting it built. The handoff to design or eng. The fight for a roadmap slot. And the hope that people grasped what you meant.
That constraint is mostly gone. AI will build almost anything you can describe, in an afternoon. Building is close to infinite.
So the bottleneck moved. When you can make anything, the scarce skill is no longer "can you make it." It is "do you know what to make, and what to leave out." Focus and relevance are the new hard part.
I did not get this from a trend piece. I watched it happen four times in one week. Twice in my own Build workshop, twice in conversations with executives who had nothing to do with it.
What happened
Over two pilot sessions, eight researchers each spent three hours turning one or more research recommendations into something a stakeholder could see, click, or use. Most arrived able to describe what they wanted, not build it. By the end, all eight had built something, and every one of them shared it during the workshop. One turned a dense, single-page diagram into a fully functional prototype. I am proud of that and proud of them!
Here’s the bigger point. Almost no one struggled with the AI's capability. They struggled with restraint.
LF said it best: scope down what you ask the AI to do, or it builds a whole product when you wanted control over a single screen.
KV made the same move with her feet, choosing the prepared project on purpose so she could focus on the skill, not the content.
And DC named the constraint from another angle: how do you keep the prototype telling the story of your recommendation, instead of chasing every detail the AI is happy to generate?
None of these is a technology problem. They are judgment problems, the kind research has always handled: deciding what matters and what to leave out.
The most-loved part of the workshop was not the AI. It was the prompt library, the set of constraints that told people where to point all that capability. That is not a coincidence. When building is nearly free, the prompt is the craft. Scoping is the craft. Knowing what to cut is the craft.
A build surfaces what a report never could
Here is the biggest aha, and it landed hardest in the second pilot. When you hand the AI a recommendation and watch its first pass, you are watching exactly what happens when you hand a written recommendation to a designer or an engineer to build. The misinterpretation. The drift. The not-quite-what-you-meant. Except now it shows up in seconds, right on the screen, instead of surfacing weeks later in a build that missed.
LF hit it live. She was turning a recommendation into a build, and the first version came back not quite in the voice its founders intended. That gap was the real finding: the intention was clear, but the product was not carrying it yet. A report would have buried that. The build made it impossible to miss and easy to fix.
That is the true superpower here. Building with AI not only allows you to bring your recommendation to life.
It shows you how your recommendation gets interpreted, immediately, so you can right-size it before a single human runs with it.
Same thesis, different sport
The same week, I had versions of this conversation with two executives in two different industries. Both wanted to talk about what to build.
One opened with a polished, multi-screen solution and asked what I thought. I stopped him. What problem are you solving? What is your evidence it is a problem? How severe is it? And who are we designing for, because I count at least six buyers here. The process does not disappear because building got easy. It matters more now, so we do not build the wrong thing faster.
The other was staring down an ocean of options. She did not want to chase 20ideas. She wanted the right one to three, and she “works best with guardrails, not a blank canvas.” So I told her what I keep telling everyone. We have more information than ever. The problem is no longer getting it. The problem is knowing which of it is worth acting on. And how, where, and when. And whether the why behind it is meaningful at all. Same thesis. Different sport. The tools changed. The discipline did not.
Repeat after me. Do not comment on the shape of a solution before you understand the problem you are trying to solve.
But the best builders say trust your gut
Fair pushback, and the best builders make it. For a true v1 in a category that does not exist yet, there is no data to lean on, and forcing it gives you an undifferentiated solution. Agreed. But look at what those builders actually do. Gather input. Prototype. Make the opinionated call. That is not skipping research. It is research informing judgment, which is what it always was. And most of us are not on a blank canvas. We are fixing a known issue or seizing an opportunity at an org with thousands of users. That is not a guess. It is an institutional knowledge signal. Trust the gut that signal built. It is how Ask Like A Pro: Together, my Curiosity Tank research partner now 29 beta tests deep, came to be. I trusted mine.
Chew on this
The arrival of AI does not make your research skills less valuable. It makes them more valuable, and it moves your work upstream.
The questions you have always been good at, what matters most and why, who is this for, what is the one thing this needs to say or do, are now the questions that separate a useful build from an impressive mess. The infinite part is cheap. The focused part is yours. Own it.
How I'll iterate
Pilots exist to surface what to improve, and these two Build Like A Pro pilots did exactly that. A few things surprised me. People came in at very different skill levels and access, from free chat tools to full builder platforms, and that spread was harder to teach to than I expected. I'll set a clearer tooling baseline up front. Pilot one asked for a quick demo at the start, so I added it for pilot two. Pilot two suggested more prep done before we met, so live time could go to building instead of reading. All fair. All going into the next version. I also added four pro tips to the prompt library, two after the first pilot, and two more after the second pilot. I assumed all were too obvious to spell out. They were not.
What's next
The pilots made one more thing clear: people want to go further. Not a gentler second pass, but depth. Iterating on a build based on how people react. Connecting a prototype to live data. More complicated builds, like a conversational interface. Grasping the difference between a prototype and what it actually takes to build for sale and scale. I had a hunch beforehand. I felt it in my gut. I'll follow it. That's shaping what comes next.
For now, I am refining Build Like A Pro and running it again in July, with less friction up front and more time to build.
Your prompt for the week has nothing to do with AI. Look at your last research recommendation and ask: if I could show my stakeholder only one fix, which one would it be? Start there. The focus is the work.
Constraints are your friend. Happy building.
Michele
