I’ve spent 10+ years designing interfaces for fintech, DeFi, and blockchain products used by millions to manage their money. For years, I handed engineers Figma files and pretended the “ready for dev” tag meant something. In reality, it meant “now we wait and hope for the best.”Every product designer faces the same wall: you can create sophisticated visuals, but if you can’t code, you can’t control the final result. That's why, for years, the product dream team looked like “PM + designer + developer.”This year's AI in Design Report, which surveyed more than 900 designers across 60+ countries, found that half of them have already pushed AI-generated code to production. I’m one of them.My name is Andrii Bondar. I’m a product design engineer at Matter Labs and founder of DecentWealth, a private iOS portfolio tracker that doesn’t require an account to start.I had a goal to build an iOS app and learned Xcode for it, but everything changed when powerful AI tools emerged.Over the 2026 New Year holidays, instead of eating steaks and cakes, I opened a Claude Code terminal. By mid-February, my first app was live on the App Store.The Trust IssueGrowing up in post-Soviet Ukraine, my parents only trusted what they could physically touch: apartments and cash under the mattress. When banks vanish overnight, and savings turn to nothing every few years, you stop believing in systems.Unlearning this took years. First through crypto, then long-term stock investing.I tried several portfolio apps. Some lacked privacy, while others assumed you were a full-time trader. My Google Sheet alternative quickly buckled under complex API connections and privacy limits. Diversified investors with stocks, crypto, and cash had no single place to see their full net worth without handing financial data to another company with a “create an account” screen.So, I built DecentWealth.Flying blind on purpose.A privacy-first app means zero analytics, session recordings, or heatmaps. Since I promised not to watch what users do, I learned to listen instead.Paid users are the best product researchers I've ever worked with. They don't send “cool app.” They send detailed, specific feedback on how they use the app:A pure black OLED theme (regular dark mode wasn't dark enough)Apple Watch features and lock screen widgetsInteractive Brokers integrationThe product covers a very specific niche, and it’s definitely not for everyone, so I see how this audience differs from the audience of neobanks or users of crypto exchanges.4 things I learned while shipping the app.Lesson 1: "Just one prompt" is the biggest bullsh*t on tech Twitter.It sells the fantasy that skill is optional. But you can't prompt your way to a design system; you have to bring one. Anyone with real product sense knows good products require experience, time, and deep research.Instead of basic prompting, treat AI as a team: ask it to take distinct roles (UX designer, CTO) and let it interview you first. Feed it unique information that agents can’t automatically scrape, like browsing forums or Similarweb competitor data, to ground its output.Lesson 2: AI made engineering easy, but marketing is still a bottleneck,A dev who just shipped their 14th app this year isn't a productivity machine. They're running from marketing. Building app #15 is easier than promoting app #3. A new app is fresh dopamine: no users to disappoint, no churn graph staring back at you. It feels like progress and looks like a portfolio.Promoting an existing product is difficult: talking to people who didn't sign up, sitting with retention numbers, and fixing the same screen for the eleventh time.Many vibecoders have gotten obsessed with the concept of taste. As a product designer, I couldn't be happier, but I deeply believe there are many technical things that still must be worked out. Paul Graham from Y Combinator says taste is the new differentiator; my co-founder Daria Volkova argued in the HackerNoon article that it's a cope. Both are right: taste gets people to stay, but it doesn't get them to show up.Here’s how I do basic product marketing using AI agents. I built a separate "CMO repo" on GitHub as a dedicated folder with:Marketing skills I useInstructions to act as a skilled CMO who knows the domain wellResearch data and scripts to collect fresh dataAnalytics data and core metricsCompetitor researchTranscripts of user callsInterview this agent like a CMO co-founder before any major update.Lesson 3: Ship messy, decide deliberately.At a recent AI meetup, founders shared what happens when execution moves too fast. Teams compressing two-week sprints into two-hour sessions risk zero time for reflection.Speed without intention is just organized chaos. Skipping tradeoffs costs weeks of debugging later. Despite that, I’m still convinced that messy execution beats a perfect strategy that was implemented too late.The “vibecoding bubble” is the most efficient business model “lab” I've ever seen. We’re watching a global, high-speed stress test of every niche imaginable. What used to take a seed round and 6 months of discovery is now being validated (or killed) in a weekend. HackerNoon, Reddit, Indie Hackers, and startup Twitter are parts of the largest open-source playbook of "what works" we've ever had.For DecentWealth, I shipped 7 versions in 9 months of the product's existence. I guess it's a good result taking into account the specifics of the fintech niche.Lesson 4: AI tools can build anything, so the job is saying no.When building costs almost nothing, everything feels like a good idea. Claude Code and Codex never tell you a feature is bad. Ask for a social feed in a private tracker, and it asks what size you want the avatars.Adding too many features creates future debt. I try to spend more time on fewer features and let users and data tell me when the app really needs the next one.For example, users wanted Interactive Brokers (IBKR) integration. Traditional apps route accounts through aggregator servers. To keep the privacy promise, I used IBKR's read-only tokens:Token stays locally on the user’s device (no iCloud sync). It can't trade or move money.Direct device-to-IBKR communication and no middleman serversClaude Code built it quickly. The hard part was deciding whether it should exist and on what terms.What's next if you're a designer still living in Figma?Designers who ship code with AI are twice as likely to feel creative, capable, and hold a higher quality bar.If coding feels like a world you're not supposed to enter, give Claude or Codex a week. Don’t build right away. Just ask questions about CSS structure and implementation. Your designs will soon reflect not only your visual skills, but also your technical authority.👍 Liked this article? Follow me on X and LinkedIn.💬 Designers who ship: what’s your experience with AI coding tools? Share your stories in the comments.