Yard

Inspiration

My long-term vision for Yard is to revive Lunch Club, and build a social search algorithm trained on one-on-one meeting notes that are tracked through the platform—but it took me enough tokens just convincing Bolt to turn this into an MPA instead of a single page app, so we still have a long ways to go!

What it does

There are three parts to Yard: the time feed, cross-resource directory, and agent-driven work bounties. The time feed is a Venmo-esque social feed of time transactions between work friends—after you get off a call with someone, upload your meeting notes (what I'm working on next) and log that time with them on Yard. Anyone you've logged time with immediately becomes a connection, and you'll see their transactions in your time feed. The cross-resource directory is an aggregated directory of profiles pulled from professional services sites like Github, Stack Overflow, Clarity, Behance, and Contra. I'd like to look into integrating with services like GLG, and perhaps Linkedin, as a way for their professionals to treat Yard as an inbound client pipeline. Finally, we have bounties. To make the platform interactive, search isn't static: we generate bounties based on AI-suggested search parameters. When you submit a bounty, we call an agent that scrapes the most relevant pre-approved sites and emails you a list of 10-12 potential candidates—next, I'll build a separate agent to handle that outreach. My thinking with bounties is that this lowers the friction to collaboration. In my H2 vision for Yard, I imagine that most users will have agents bounty hunt for them, so I'm also thinking about how to make bounties as LLM-legible as possible.

How we built it

I brainstormed v0 with v0, designed a landing page on Figma and coded it with Magic Patterns, and then fed Bolt the PRD and landing page code and we workshopped the rest together. Bolt suggested some strong UX decisions, like changing email invites to invite links.

Challenges I ran into

As a non-technical builder, I found myself held back when Bolt made product decisions that I couldn't contribute to. The SPA vs. MPA debate was a big one, and trying to prompt Bolt through all the hyperniche auth/onboarding flows I was envisioning was a nightmare. It really made me think a lot about Guillermo's allocation economy ethos, and there was a lot of turning edge functions into prompts that I could ask Claude for help with, and relaying those answers back to Bolt.

What I learned

All in all, a very fun initiation into vibe coding! When I opened my terminal for the first time last year to build Digest.town, it took me four months, and a tutor(!!) to get to the same level of janky-finality with the app as it took me a week with Bolt to get Yard to. Had I not had a tutor, I could see it taking me a lot longer to get to the same place with Bolt; the learning curve that tutorship seems to have flattened for me was developing the language for efficient prompting. That said, where I'm finding myself limited in prompting capacity now is nitpicking the right bugs—most of the time (and token) sink I'm experiencing is in not being able to provide more direction during debugging.

Built With

• bolt
• magic-patterns
• supabase
• v0