From a Week of Learning to an Hour of Building — This Is What AI Progress Actually Looks Like
Seven months ago, I spent an entire week using AI to teach me how to build a software system from scratch.
Not using it to do the work — using it to teach me the work. I had this idea: what if you could take a company, identify its MVP, benchmark it against competitors and parallel players in the industry, find where each one falls short, find where one company quietly fills the gaps of another, and layer on top of that a scraping system pulling from review platforms — Trustpilot, Help Scout, G2 — to surface real FAQs, sentiment trends, and ratings? Feed all of that back into the business model. Sharpen the focus. Define the target market with actual signal, not assumption.
It took a week. A full week of back-and-forth, learning what I didn't know, asking the right questions, rebuilding when something didn't work, and slowly connecting the pieces into something coherent. I was proud of it. I still am.
Yesterday, I built an entire prospecting app in under an hour. Zero lines of code written by me. Zero.
The app identifies ICPs, actively searches for similar companies, and learns from each result — adjusting what to bring more of, what to pull back on, and building a prospecting process that gets smarter as you use it.
But it doesn't stop at finding companies. It builds the entire outreach strategy around them. It surfaces the best contacts at each company based on publicly available information, identifies their most relevant use case for your product or service, and drafts a personalized email and LinkedIn message — not templated, actually personalized. Think: "You've been operating since 2006, you've doubled your inventory and tripled your headcount — here's why I think there's something here worth a conversation." Real data, pulled and applied in context.
From there, it categorizes every lead into hot, warm, or cold, and builds a sequencing program that tracks open emails, responses, and time since last interaction — so nothing falls through the cracks and follow-ups are timed with intention rather than gut feel. And layered on top of all of that is a full pipeline view where I can zoom into hyper-specific markets: companies in Denver with $1M+ in annual revenue, a public web presence, and a website updated every six to eight months. The kind of targeting that used to require a research analyst and a week of work.
Something I would have spent days on seven months ago, I shipped before lunch.
That's not a flex. That's a data point. And it says something worth sitting with.
The gap between those two moments isn't really about AI getting better (though it has). It's about me getting better at using it. And there's a real distinction there that I think gets lost in most conversations about AI tools.
Everyone's rushing to find the right tool. Which platform, which model, which subscription. But the tools are only as useful as your ability to actually use them — and I don't mean technically. I mean conceptually. The ability to take a messy, half-formed idea and translate it into a conversation that produces something real. That's the skill. That's what makes the difference between someone who gets a mediocre output and someone who builds an app in an hour.
There's a term for it: prompt engineering. And I know that phrase sounds a little clinical, a little overhyped — but at its core it just means being able to put your ideas and your thinking into a real, thoughtful dialogue. Knowing how to frame context. Knowing what to push back on. Knowing when to go deeper versus when to start fresh. It's less about commands and more about conversation.
I've been getting to watch this play out firsthand at CARU. I've started sharing some of what I've been building with the team — not as a training, just as a "hey, look what this can do" — and watching people develop their own instincts in real time has been genuinely one of the more interesting parts of my past few months.
People are landing on preferences. Some are gravitating toward ChatGPT, some toward Claude. Some are using both for different things, which honestly makes the most sense. And while we're still early — entry-level in a lot of ways — you can already see the seeds of something more powerful forming. The moment someone stops asking "what tool should I use" and starts asking "how do I get this specific tool to do exactly what I need" — that's the shift. That's when it stops being a novelty and starts being leverage.
Seven months from now, I'm curious what the version of this post looks like. What I'll build in ten minutes that took an hour today. What my team will be capable of that feels ambitious right now.
But more than that — I'm curious about the people who never make the jump. Not because they don't have access to the tools, but because they never develop the skill to use them with intention.
So here's the question I keep coming back to: is prompt engineering something you can teach, or is it something you have to earn through the doing?
