@RobertJBye Git interactions with worktrees, merging, creating PRs, seems messy in Code. Feels like the UI doesn't really support the workflow. I mostly rely on chat to separate out changes and migrations
Validating/citing the content was important for me, hence storing page text. Another layer... when you're not sure which page to fetch: crawl a handful, keyword rank works fine, then run a separate jev call to pick which fetched page is the one you need.
Jev is great at turning messy sources into relatively clean, structured datasets. Many approaches! But a simple pipeline for web content: fetch with Cloudflare Browser Run → store the text → route typed questions to jev → store the output.
@johnrush Have all the micro saas opportunities really been eaten? Feels like there are many dated-and-neglected but still used apps sitting out there that are vulnerable to a solo founder nuking them on features and pricing with some ambition and Astra
@gregisenberg Good points! Capability is there. Even Sonnet or Opus can drive KiCad. Design and validation is still tricky to one shot. JLCPCB has min pieces of 5, so < $100 with full assembly is more like $250-$500. Still super cheap. https://t.co/ciNlKf5py8
I went through the process of taking a hardware idea all the way from Claude prompts to a custom board.
Totally possible, but way easier to start with off the shelf parts (like the Build In Public gadgets list). Sharing my 15 min learnings recap: https://t.co/0GBoSphDRb
Working with a few SMBs. The owners that are tech aware do see some AI hype. The easier wins mainly reduce human time... routine tasks where a good skill can help. Compounding leverage from AI in SMBs feels more dependent on data architecture and reorienting processes though. Much slower to achieve that when margins and resources are thin.
@coreyganim Great insight, thanks for sharing. Are you finding 4 hours is a sweet spot for balancing attention and live building? Curious if shorter sessions (1-2hrs, maybe include lunch) would be well received.
@supabase Have a few Supabase projects in flight. 👋 Great platform. Building an enterprise-grade HR performance appraisal system... needs to handle 40K staff. Deep into a fintech SaaS platform as well, and my AI company is Supabase for the backend + client builds.
I went through the process of taking a hardware idea all the way from Claude prompts to a custom board.
Totally possible, but way easier to start with off the shelf parts (like the Build In Public gadgets list). Sharing my 15 min learnings recap: https://t.co/0GBoSphDRb
@noahfradin@BradButner@Timothy_Laurent Hey, Noah! Thanks for sharing this. We'd be interested to try a local classifier model as well - our team routes requests through LiteLLM.
Turns out Phoenix, AZ and the Forest Service needed a Tonto Pass locator tool. Fun to watch an idea become something people use every weekend. Hope it's made a few adventures a bit easier!
Quick weekend Claude Code project. Some great hikes and recreation areas in Arizona require a Tonto pass, but for some reason no map existed for finding one.
- maps: mapcn (thanks @sainianmol16!)
- autocomplete: photon
- geolocating: Nominatim
https://t.co/5Rax2jF0z6
@pbakaus Great to know! Thanks, checked diffs closer and looks great. If project isn’t using Tailwind, appears /normalize can create a design token system from scratch. Really neat. 👍
@lucknite Crazy. Impressively capable but at the cost of insane exposure unless you truly know what you’re doing. Feels like the average user shouldn’t be touching this without heavy isolation.