Copia is for buying food from your neighbors, the people who grow it. It’s built the way a farmers market feels: calm and personal.
At a glance
- Role
- Founder · sole designer & builder
- Build
- 2021 (as EDEN) · rebuilt 2026 in 17 days, directing AI
- Tools
- Figma, Illustrator, Photoshop, Procreate, Higgsfield AI, and Claude Code
- Stack
- Next.js · TypeScript · Supabase · Mapbox · Claude · Vercel
- Shipped
- 9 seller surfaces · full buyer flow · 1 production AI feature
- Tested
- 2021 prototype with the growers I’d interviewed, at the markets; 2026 rebuild with first-time users, watching where the flow held and where it snagged.
- Live
- copia-virid.vercel.app
The short version
I founded Copia in my MBA in 2021, as EDEN. We pitched VCs and were offered $1.2M. I declined: I was the only founder all-in, and I had no technical co-founder to build with. By 2026, AI had removed that barrier, so I went back and built what I’d envisioned all along, alone, directing AI through every screen and every sentence. Seventeen build days to a live, working prototype.
Local food runs on a once-a-week clock.
Farmers markets open for a few hours on a Saturday, the only window most people have to buy produce from the person who grew it, and the only window growers have to sell it. Step outside that window and it breaks down: buyers can’t find trustworthy local growers, and growers can’t find customers. It started during the 2020 food shortages, when the gap between wanting to buy local and actually getting it was at its widest. Copia set out to unbottleneck both sides: an always-open marketplace for buying and selling fresh local produce, any day of the week.
What 47 interviews said (2021)
And the surprise: people value local over organic. Some actively avoid the organic label.
Alongside the interviews, I mapped every job, pain, and gain on a value-proposition canvas, sized up the competition, and charted the ecosystem Copia would live in.
Decisions
2021 next to 2026.
2021 proved the concept; 2026 gave it a voice. The early prototype carried the whole thesis (local sellers, mapped, transacting in one place), and the research behind every screen here came from it. But it looked like everything else. The rebuild kept the soul and replaced the surface: editorial type, a weekly almanac in place of a feed, calm in place of clutter. Same idea, finally made with care.
One calm page instead of the checkout wizard.
The first version put one grower per checkout, behind big product images, and it confused everyone who tested it. At a market, people happily wander stall to stall; in an app, nobody wants to. They expect a grocery cart, not a market stall: one basket, several growers, settled up all at once. So I rebuilt it around that: every grower group still picks pickup or delivery, deliveries consolidate into one courier run with one fee, and the whole plan reads back as a single sentence. “You’ll collect the heirloom tomatoes from Mira at the Mueller market Saturday morning.”
A calendar that refuses the month grid.
Nobody buys squash more than a month out, so the seller calendar is an agenda bounded at four weeks. The genre default would have fought how fresh food is actually bought.
An almanac on the phone, analytics on the desk.
Sellers wanted data, just not the same data on every screen. On the phone, the almanac wins: a plain-language read of what happened and what to do Saturday, no charting library in sight. The long-term view (trend lines and season-to-season overview) is reserved for the desktop or iPad workstation, where there’s room to study it. Same numbers, sorted by what the moment calls for.
Agentic AI as a letter, not a chatbot.
The Coach reads your plot (orders, listings, the forecast, what neighbors are searching) and writes you a weekly note in the product’s own voice. It’s AI-written, and the letter says so.
Where AI isn’t.
When two growers share a market day, the page suggests one trip. That’s geometry, not a model. Knowing where not to use AI is part of AI design.
Two surfaces, two contexts.
Buyers and growers don’t share a job. Buyers shop in spare minutes, so the buyer experience is iPhone-first. Growers work both ways (a quick check on the phone in the field), but to actually run the business, they told me they’d rather be on a real screen. So the seller tools live in both places: the phone for quick moves, and a full desktop workstation (dashboard, orders, listings, calendar, the Coach) for the work that needs room. Same brand, same calm, the right shape for each.
How it was built
Before any of it was code, it was wireframes. Copia’s first life was 2021, as EDEN. I designed the branding, the onboarding, and a full low-fidelity buyer flow, sketched and iterated screen by screen in Figma. That groundwork is what made the 2026 rebuild fast: I wasn’t inventing the product while directing the build, I was executing a plan I’d already drawn.
Then, in 2026, I directed Claude Code the way a creative director runs a studio: a written design directive with a NEVER list banning the generic-app vocabulary, a paste-ready brief for each build day, and every word of copy reviewed before merge. Every brief is committed alongside the code. One engineering proof: the Coach’s cold first paint went from 16 seconds to 0.26, by fetching the letter behind the brand’s leaf-draw animation.
Reflection
The hardest part was refusing: the month grid, the progress bars, the trust badges, the sparkle. Next would be real payments (Stripe for cards; USDC settlement matters most where the local currency doesn’t hold), a smoother on-ramp for the buyers who, by the end of an interview, wanted to grow and sell too, and the long thread from my 2020 notes: robot-tended backyards removing the last labor barrier between “I’d like to sell some tomatoes” and “I am, in fact, selling tomatoes.”
On the home screen
Add to Home Screen from Safari and the leaf joins the other apps. The real iPhone status bars are the point: a genuine install, not a mockup of one.
Try it
Open it in Safari on an iPhone. Put three things from three growers in your basket and settle up. For the full effect, add it to your home screen: the wordmark assembles itself on the way in.
