It started as a webpage for one family vacation.

Three nudges into vibe coding

It took three nudges to get me into vibe coding. First, in early February, a colleague went on a ski trip, shared a lift ride with an Apple executive, and came back asking, "Have you heard of Claude Code?" Then, in mid-February, I read Matt Shumer's article Something Big is Happening. Finally, in late February, a close friend texted me, "I've taken up vibe coding. It's addicting. Soon you'll be writing code for everything!"

I decided to start small by asking AI to build a personal webpage centered on my career. In less than ten minutes, it produced a polished site with my headshot, bio, and links to my academic publications organized by category. Around the same time, I'd listened to an episode of The Daily where they used AI to create a simple game during the broadcast. That inspired me to ask AI to build a game about accelerating a proton in a cyclotron, tying into my work as a physicist. Five minutes later, I had a version that was about 90% complete and just needed a bit more direction. That helped me understand both the power of vibe coding and its basic workflow: prompt, review, iterate.

A webpage for one vacation

Fast forward a few weeks and I decided, as another experiment, to build a webpage for our family's first trip to Hawaii. It started as a simple text-based site to keep track of our itinerary, but AI kept making it easy to add more. During the trip, I enhanced the page with world clocks, an interactive map, an eye-spy game, live weather, and eventually a feature that let our group upload photos to each day's itinerary via Cloudinary. Some of those updates I made right from my phone while waiting in the TSA line.

Another lesson learned: AI agents don't just do the work — they generate ideas, roadmaps, and possibilities much faster than humans can execute them. They expose an "infinite backlog": there's always more to do. (That framing is borrowed from The AI Daily Brief podcast — "Why Agents Make Every Job a Startup.")

Once we were in Hawaii, everyone kept pulling up the page to check the next day's plans and upload their photos. Our family back home loved being able to follow along too. Half-way through the trip, the consensus was that this could turn into something more — so we spent a few days having fun brainstorming names for the potential new app: SherpaSnap, SherpaDoodle, Cardfly, Ohana Deck, Kintinerary, and finally Kintinery — a portmanteau of kin (family and friends) and itinerary. I also used AI to draft an initial business plan: product description, target market, competitive landscape, business model, marketing strategy, technology stack, estimated operating cost, and recommended next steps.

The pivot to a real product

The first movement along that path started during the trip itself: I set up the AWS Management Console and migrated photo storage from Cloudinary to Amazon S3 for a more robust image backend. I decided it would be easiest to keep the entire stack within AWS, so I took time to understand the core services — Cognito for authentication, Route 53 and CloudFront for distribution, DynamoDB for the database, Lambda for application logic, and later Bedrock for AI inference. Fortunately, AWS provides a command-line interface that Claude Code could access. Once the infrastructure was in place, AI was able to take on much of the deployment work.

When we got home, I bought Kintinery.com and began building the web version, carrying over much of the functionality from the original site. I also spent significant time developing the brand identity — the color palette, typography, taglines, component patterns, and the logo. AI was especially helpful for generating ideas, producing quick mockups for review, and creating design documentation. For the logo I used Gemini to generate images and iterate through concepts; eventually I chose a banyan tree, which reminded me of Hawaii and the app's origin, then refined it in Photoshop into multiple color and wordmark versions.

With the design system established and the web app already in progress, I built a landing page — likely the first thing potential customers see. The initial version had the right elements but lacked the feel of a modern site. I briefly subscribed to Figma to explore a stronger design, and it helped — but I didn't want to host through Figma's tools, so I fed Figma's output into Claude Code to see whether it could recreate the result. It correctly identified React as the missing piece for an interactive UI, and using that approach I redesigned the landing page into the version that's now live.

A key step was integrating payments through Stripe — creating products, preparing documentation, and a lot of testing, with AI guiding me through. (I took screenshots along the way so I could share them with AI whenever I had a question — a habit that paid off again and again.)

iOS in a single day

Once the web app was functional, I was eager to build the iOS version. As a longtime PC user without access to Xcode, I had to choose between cloud-based Mac services, cross-platform frameworks, or buying a Mac. To reduce friction I bought a 13-inch MacBook Air — a great decision: it made iOS development much smoother, and I love that I can take a screenshot on my iPhone while running the app and have it appear instantly on the Mac, ready to hand to the AI for review.

With the MacBook in hand, I created the initial iOS scaffold in a single day, then spent the next month adding features, testing, and fixing bugs — the same prompt-review-iterate loop. I tested in the Simulator, on my own device, and later via TestFlight. Pricing the two products — Pro for a Trip and Pro Annual — took real thought; I used AI to model scenarios and took a deep dive into estimating inference costs and controlling them through model selection (Haiku vs. Sonnet), prompt caching, and sensible usage limits on premium features.

For the release I needed professional screenshots, so I used appscreens.com to create a polished, on-brand set, then worked through App Store Connect for the first submission — AI helping with Q&A and App Store Optimization. My first two builds were rejected and needed minor fixes over a day or two; the third was a success. Seeing the Kintinery app appear in the App Store, just like any other app you might search for, was both rewarding and a little surreal.

Going legit and first real users

Around the same time I applied for the Apple Small Business Program (still pending two months later) and incorporated as Kintinery LLC. I applied for a DUNS number — annoyingly complicated — and spent a week trying to get an EIN from the IRS, which required a fax plus a couple of follow-up calls (the trick: be first in line when they open, or risk a multi-hour wait). With that in place I opened a business checking account and migrated my developer account from individual to organization — about a week that blocked any further App Store releases, which was frustrating while I was chasing bugs and trying to ship fixes.

My first real users were my in-laws, on a multi-week vacation to Europe. They built their itinerary by feeding in emails and documents from their travel agent, and while traveling they were great about uploading photos — which the whole family back home enjoyed as a way to follow the adventure. When they got home, I spent about an hour listening to my mother-in-law's feedback, which was golden. Several features and fixes came directly from her experience.

Security audits and early Android build

Periodically I'd ask AI to run deep security and performance audits, and I'd routinely ask about vulnerabilities I saw discussed on Reddit. Most of this was done with Opus on max settings, but I also ran it through the famously capable Fable 5 model during its brief availability in early June. Fable 5 flagged only a few minor things, which gave me confidence the app was in good shape — and since the entire stack runs through AWS, the infrastructure is robust as well.

I also used Fable 5 to start the Android project, using the iOS repository as a reference. It produced a six-phase plan and implemented much of it on its own — independently testing by driving the emulator and taking its own screenshots. AI got me 90% of the way; the final 10% (validation, Google Play and Cloud Console setup, Firebase) was still a significant lift. Android distribution is far more involved than iOS — the Play Store even requires video review for certain features, so I record screen captures and upload them to YouTube. I'm hoping to push an early release soon.

Worth it

I believe in the product and that it fills a real niche — but even if the app isn't a financial success, it's been an incredibly fun and instructive project. I've learned how to vibe code, how to establish a business, how to integrate every component of the technology stack needed to run a SaaS, and how to manage the app release process — and now I'm learning what it takes to market and distribute a product. If you or your family have a trip on the horizon, I'd love for you to give Kintinery a try. Your feedback would mean a lot.