What should you build when you can build anything?

Read that quote again. Ol’ Eugene had a point, and it’s more true now than ever with AI dominating our lives.

I was lurking in a community earlier this morning and stumbled on this post:

You get the idea. This person, while very well intentioned, built an impressive machine: sophisticated dashboards, automations, and a comprehensive ICP study tailored to their company. And they didn’t actually use it.

The most sophisticated tool can still sit unused if nobody inside your team is asking the right questions.

AI makes it incredibly easy for all of us to lock ourselves in a room, build and build, and mistake all that output for learning.

That’s the problem.

More data and more tools aren’t signals of real progress or insight. This is the new reality.

We’re rushing to build stuff, launch products, and iterate on marketing and messaging without knowing whether any of it will help. Just because we can.

But the fact that you can doesn’t mean you should.

Getting more stuff out there might feel productive. Launching, however, will never tell you on its own whether that thing should exist, what customers care about, or how you should position it so it stands out.

There’s only one way to figure that out: listen to your market.

Knowing your customers is a filter. It helps you decide which problems are real, what language people already use, and whether your intended positioning and messaging are actually working. Knowing yourself and your point of view sends you in a direction. Knowing your customers and how they perceive you tells you whether you should keep going.

​Hiten Shah has a great point about deciding whether to build something when you can literally build anything. He frames it this way: “You are basically looking for machinery that has already proven it can change human behavior. Then you figure out whether that machinery can help with a behavior that matters inside your own product.”

In other words, don’t copy a feature because its metrics look good or run wild with whatever you can now prototype. Look underneath it. Figure out why and how people use something that already works, then ask whether that same behavior matters inside your product.

Behind this approach there’s still a deep understanding of how humans think and act.

​Peep Laja has a great point too: measure how buyers perceive your brand and whether they remember you, not just their clicks and conversions.

Figuring out whether buyers in your ICP are familiar with your company and can accurately describe what you do matters now, probably more than before, because buyers can also meet your positioning in AI-generated answers.

My point is that a dataset isn’t a signal by itself. A signal depends on a question, a decision, and a bar for taking action.

When I have to make an important decision and understand what data I need, I always go back to the OODA loop:

• Observe: What’s happening right now, and what are we deciding? Say a B2B SaaS company has expanded beyond its original use case. The decision might be whether its homepage should still lead with the old point solution or the broader workflow it now supports.
• Orient: What’s blocking that decision, and what do we need to learn? Maybe the team knows the product has changed but not whether buyers understand the new category, trust the bigger promise, or still come in for the original job.
• Decide: What evidence will help us choose, and how will we get it? Review sales calls, interview recent customers and lost prospects, compare the language each group uses, then decide which message deserves the lead.
• Act: Who owns the decision, and what changes next? One owner updates the homepage and sales story, then watches whether buyer questions, objections, and qualified conversations change.

This stops you from acting before you actually know what to act on and how.

And when it comes to what evidence to gather, customer research is still the answer. Real people are the standard. If they’re genuinely unavailable, synthetic ICPs can help you map possible routes and find better questions.

Customer research gives you what siloed building never will: what your ICP thinks the problem is, how they talk about your company, category, and competitors, which trade-offs they’re making, and what gives them confidence—or takes it away.

Just don’t ask them what to build.

You’re studying their situation, language, past behavior, and decision-making process, not outsourcing product management to them.

You’re getting at what to build, by understanding them and their needs.

If you get to speak with prospects or customers even once (better than nothing), I want to leave you with the simplest three-question framework I use. Call it a customer perception check.

1. Recognition — Have they heard of you? Does the brand come to mind at all? Ask: “When you think about [buying situation/category], which companies come to mind?”
2. Description — Can they explain what you do? Can buyers describe your category, offer, and ICP accurately in their own words? Ask: “What do you think [company] helps people do, and who is it for?”
3. Association — What do they connect you with? Do they spontaneously link your company to the particular problem, advantage, belief, or buying situation you want to own? Ask: “What is the first problem, advantage, or idea you connect with [company]?”

In an ideal scenario, you want to be easily recognized, clearly described, and associated with the right thing.

Aim to get as high as you can on that ladder.

Pro tip: ask both prospects AND customers. The difference between their answers can expose the gap between what your messaging promises and what the experience actually delivers. Then you can work to align them.

I’m the first proponent of AI. Hell, I’ve spent the past four months building systems, launching workflows, and playing with agents like a maniac.

Yet before I rework the same area again, I also have to wait patiently for the market to speak to me.

I listen. Then I rebuild and relaunch.

When building gets cheaper, the most expensive mistake is shipping, measuring, and optimizing something nobody needed, or becoming known for something you aren’t and never wanted to be.

Discovery

Synthetic research is a map, not the territory

I’ve written about synthetic research before, but John Horton’s piece on AI social simulations gave me the clearest metaphor yet for where it helps. He compares a simulation to a map and most analytics to a trip report. A map can be incomplete, distorted, and still help you compare possible routes before you travel. You just don’t confuse it with the terrain.

That’s how I’d use synthetic ICPs: not as an accurate replica of human psychology or proof of what customers will do, but as a way to explore directions, find missing questions, and decide what deserves a real-world test.

Adaptability, not endless optimization

​Jordi Visser’s essay on HRV, happiness, and AI is a long-ish read, but the idea is simple and very relevant today: adaptability matters more than strength or one perfect routine. His line that “the frameworks are transferable; the tactics are personal” is especially useful when every expert, dashboard, and AI system can give you another thing to optimize.

For those of us dealing with constant information abundance, stress, and the sense that we should always be doing more, it’s a useful reminder: the goal isn’t to absorb and optimize everything but to build a system that helps us find the balance.

It made me want to learn more about HRV!

Resonance

• Joanna Wiebe, The Copyselling System