How to bring a Product Manager dream into reality with AI

You might ask: what is a PM’s dream? Sure, the obvious answer is business success. But I’m not going to bore you with yet another step‑by‑step guide on how to find a use case, run market research, do pricing exercises, prototype, collect user feedback, and track KPIs for AI business success. There are plenty of courses and YouTube videos that do that very well, or you can enroll to a MBA program.

The dream here is bringing something into the world with the magic of AI even if you don’t have much technical experience. AI gives you an incredible source of power so you can actually chase those dreams and turn them into reality.

I felt fulfilling a couple of weeks ago when I teamed up with a few of my talented engineering peers at a hackathon. We coded something into existence, it worked, we won a prize, and it felt so good, just to be clear, we didn’t use AI to write the code.

I feel mustache also played a role in this.

For many of us who work as product managers, the feeling of making something end‑to‑end is deeply empowering. It unlocks capabilities in you that you didn’t know were there. In this short write‑up, I’m going to lay out how I chase those dreams in a technical‑ish, tactical way.

I’m going to skip all the “how to brainstorm features with AI” steps, that’s for another post. Here, I want to focus on how to turn ideas into real code.

Choose Your Magic Wand

There are so many agentic coding tools out there right now that it feels like the cereal aisle in a supermarket: a million options, all promising greatness, and instant regret with about 90% of them once you open the box.

Broadly, there are 3 main choices fore coding with AI. You can use Agentic IDE which are mostly VS code wrappers powered by AI for coding, going all‑in on “one‑prompt app builder” tools (aka your “core vibe coding” tools), or if you want to feel nerdy vibe, CLI‑based coders are built for you.

the last two are basically two sides of the AI coding tools spectrum. I’ve personally found IDE tools the most useful, for a few reasons:

1. You share a language with your dev peers

Whether it’s an engineering colleague or another technical partner, you need a common language. Using tools like Windsurf (my go‑to, introduced to me by my brother‑in‑law on a trip when I got bored and wanted to do something different), Cursor (barely used but I’ve heard lots of good things), VS Code + Copilot (I use it from time to time for troubleshooting), and others puts you closer to their world.

You’re working in files, branches, terminals, not just a magical no‑code canvas. That makes collaboration so much smoother.

Hot out of the oven: Google just released their own IDE while I was polishing this content called Antigravity, which looks more like an internal project name, to be honest, than a real product name.

I haven’t tested it myself yet, but I’ve heard that it comes with free credits for Gemini 3, so it’s definitely worth checking out.

2. Your models are not limited

Agentic IDE tools usually let you choose from different models. You can experiment with capabilities, pick what suits you best, and keep up with new releases as they’re added. There are often free tiers or promos too, which matters if, like me, you occasionally get into trouble at home for spending too much on AI stuff.

Personal preference: Claude models give me fewer heartburns and reduce the number of times I angrily type, “everything is broken, fix it all, make no mistakes.”

3. You accidentally learn version control

I barely knew anything about version control and the power of Git when I started. I still don’t know a lot, but I’ve learned so much just by using these tools.

It gives you both real power (rollback, branching, collaboration) and credibility when you talk to your engineering peers or very technical customers. You don’t have to be a Git wizard, but knowing the basics changes the conversation.

Choose the Code Architecture First (You’ll Thank Me Later)

Architecture looks like an alien language at first for many PMs (it did for me), but you can learn a ton just by chatting with AI about it.

Prompt AI with the use cases you want to build and your expectations for features and capabilities. For example:

“I want to build an AI‑based love and kindness spreader app. It should be secure, easy to deploy, handle 1000 users concurrently with no issues, be budget‑friendly to host, and not too complicated. Give me an end‑to‑end architecture from UI to backend and database.”

(Yes, I just slid in KindnessSender.com very casually. Pretend you didn’t notice.)

Even if you’re not familiar with a 3‑tier application model (UI, backend, database), don’t worry, your AI friend will explain it.

Then ask it to design the architecture and create a design principles document you can reference during coding to it time to time, so the model doesn’t drift as your codebase grows.

Pro tip: Use any “deep research” capability your favorite AI has for this step. The answer will be verbose, but it’s worth it. That document becomes your north star when you’re lost in a sea of files and error logs.

Care About the Language, But Don’t Get Hung Up on It

You may have touched programming languages in school or elsewhere and feel comfortable with one or two. That’s great, but don’t overthink it. For the UI, you don’t have that many options: JavaScript or TypeScript are the usual go-to’s and, for the backend, go with whatever AI suggests that fits your needs (Node, Python, Go, etc.).

Be aware that you will be judged by some passionate engineers for your choices, but hey, you’re chasing a dream, and there are many dream‑killers out there in the world, ignore them and move on.

Also, for some reason AI loves Tailwind CSS and gradient pink‑purple backgrounds. If you don’t mind people instantly recognizing that your creation is “an AI‑made tool,” just roll with it. Otherwise, you’ll have to be a bit more opinionated.

UI Is Where AI Disappoints You (But There’s a Way)

AI often often doesn’t really get what you want until you over‑explain it.

You basically have to do allofeeding like it’s your lovely baby bird: you have to chop the problem, cook it, chew it, almost swallow it, then regurgitate it nicely back to AI.

Disclaimer: I didn’t even know the meaning of “allofeeding” until I googled “phrase for birds feeding digested food to their chicks.”

Here’s how to get better UI results:

1. Find your dream UI and layout online. Just search for what you want: “clean dashboard UI,” “minimal landing page,” “chat interface,” etc. Show AI concrete examples.
2. Create a theme file. Ask AI to generate a theme file (colors, spacing, typography tokens) and tweak it with your own preferences. Pro tip: avoid falling in love with “rainbow everything” unless you want your app to look like a crayon box exploded.
3. Create a shared styles/typography file. Ask AI to define consistent text styles and component spacing so everything looks intentional instead of random Franken‑UI.

Even after all this, when you keep referencing those theme and style files, your agentic coder will still occasionally give you a dropdown with radio buttons in its list covered by a checkbox, three different font sizes on one card, or a layout that looks fine in its head and awful in your browser

When that happens, follow these steps:

• Be patient.
• Take a couple of deep breaths.
• Go for a walk. Touch some grass.
• Come back and write: “Fix this layout in this way. Make no mistakes.”
• Attach a screenshot or even an ugly PowerPoint sketch of what you actually want.
• Repeat until it clicks.

Don’t Be an HTTP Type of Person

Security is that thing that keep you on your toes from time to time but it is like the boiled turnip your mom made you eat when you were sick: the taste sucks, but it’s good for you.

Your AI friend shines and fails here, because you ask it for security guidelines, and it gives you a bank‑vault‑level security design, then you ask the agentic coder to implement it, and it forgets half of what it said. Your app breaks a lot, don’t panic and don’t push to production on Fridays (that’s a developers’ insider joke, don’t worry if you didn’t know, I also learned it recently.)

When that happens, yes, it’s time for another mini meditation session. Then you go back and, paste logs, snippets of failing code, config files into the prompt until the pieces match the original security plan and the app runs.

You don’t have to be a security engineer, but do not ship everything over HTTP with hard‑coded keys and hope for the best.

Become a Log Paster

AI will drive you crazy when it’s writing code for you. Make sure you have: A) A decent therapist, B) Debug logs. Good logging can easily cut in half the number of times you have to say, “It’s broken, fix it.”

Ask your agentic coder to add meaningful logs to your backend, log errors with enough context, and print out variables where things often go wrong. Then, when something fails, copy‑paste the logs from your console into the prompt, add a sentence like: “Here are the logs, please diagnose and fix.”, and let it iterate. You become less of a “panicked bug reporter” and more of a “log‑driven debugger,” even if AI is doing most of the actual thinking. when you’re ready to publish in prod, you can ask AI to remove debug level logs from the code.

At the End of the Day, It’s Just 0s and 1s. You Are the Spirit.

This all takes time, but not as much as you might think. If it were a long, painful learning curve, I would have quit tech and started a tomato farm a long time ago. I’m not a “study for years” person. I have plenty of unfinished Udemy courses where I only watched the first five minutes and switched my brain off. You will face lots of first‑time disappointments with anything new you ask AI to build. That’s normal, keep going anyway, Compliment AI with a “you’re a good boy” when the code works, it might even wag its tail by giving lots of em dashes. Or threaten it with “If you mess this up again, I’m deleting you.” Just be cautious, when AI and robots take over the world, they might take revenge.

Most importantly: don’t forget your own taste. This is your dream. You want to see it fulfilled in a way that feels right to you. AI gives you almost unlimited power to chase your ideas, but you still have to take the first step, show up consistently, and befriend AI.

Our generations are categorized by the era we were born to like: Boomers, Gen X, Millennials (please, no need for a standing ovation), Gen Z, Alpha, etc.

But I’d bet that a hundred years from now, we might all be known as Dreamers powered by AI.