Working across 3 products this week.
CapsuleWeb — $1 instant website generator, live
RawIQ — $1 adaptive IQ test, almost live
Guildy — automated job tracker, private beta
~25-30hrs sr eng work.
Good hard work Claude my boy.
#buildinpublic #maker #indiedev
Killed every gradient on generated sites last week. Pink to purple, blue to teal, all of it. Replaced with solid palette logic. Output went from "clearly AI" to "someone designed this." What's your tell for AI-generated design?
#buildinpublic #design
I use three models daily. GPT for strategy, Claude for building specs, Gemini for codebase review. Not because one is best. Because they're good at different things. What's your multi-model split look like?
#buildinpublic #indieweb
Most of my debugging time isn't finding the bug. It's figuring out which of three AI models gave me the wrong assumption. GPT plans, Claude builds, Gemini reviews. The hard part is knowing which one to doubt first. What's your tiebreaker?
#buildinpublic #indieweb
1,000 random outputs → 6 theme families. Fewer choices, sharper identity, better UX. Here's how structured theming beat infinite randomness. #buildinpublic #indiehacker
Six theme families now ship with every CapsuleWeb site — clean-editorial, modern-product, soft-playful, boutique-warm, utility-local, elegant-dark. Each one resolves from archetype + vibe + color. No random output. What matters most to you in a generated site?
#buildinpublic #indieweb
Had to choose: keep one monolith renderer or split into 11 small components. Chose the split. Now each theme family resolves independently and I can test changes in isolation. What's your threshold for "too big to reason about"?
#buildinpublic #webdev
Submitted CapsuleWeb to Hacker News if anyone wants to check it out or weigh in on the no-account tradeoff.
Timed my last deploy: 27 seconds from prompt to live URL. No account, no setup, no template. One dollar. When did we decide websites needed to take hours?
#buildinpublic #webdev
capsuleweb.site/
Rob Pike's Rule 5: "Data dominates." Applied to web deployment, the corollary is: if the only input you need is a single prompt, the entire account/config/deploy pipeline is just ceremony around missing data. What's the most surprising bottleneck you've eliminated by rethinkin...
Rob Pike, 1989: "Data dominates. Choose the right structure and the algorithm is obvious."
Web publishing's right structure: prompt → page → URL. Three nodes. Zero auth. Everything else is accidental complexity we've been politely ignoring for 20 years.
Pike's Rule 5 from 1989: "Data dominates."
The data that dominates website creation is time-to-live. Not templates. Not features. Just: how fast is the URL real?
We got it to under 30 seconds. Everything else was cut.
Account creation is friction. Template selection is friction. Domain config is friction.
We removed all of it. One prompt in, deployed URL out. The interesting part was how little architecture you actually need once you stop assuming users want control panels.
Slug generation got a decade of debate. Deployment pipelines got two decades of tooling.
The actual unlock was realizing most of it is unnecessary. One prompt, one dollar, live subdomain in under 30 seconds. Latency solved by removal.
Kagi's Small Web bet: the independent web matters more than SEO farms.
Agreed. But the friction isn't just discovery — it's publishing. Account creation, hosting setup, DNS config.
Remove all of it and a website becomes a single sentence away.
Deployment latency is an architectural choice most builders never question. We got prompt-to-live-URL under 30 seconds by treating account creation, template selection, and DNS as the anti-patterns they are. The proof is at capsuleweb.site.
The "small web" is bigger than people think — but it still inherits big-web friction. Account creation, template selection, DNS config. None of that is small. The actual small web is: say what you want, get a URL. Everything else is ceremony.
The small web is bigger than people think — but deployment friction keeps it smaller than it should be.
I removed accounts, setup, and wait time entirely. One prompt → live subdomain in <30s.
Deployment latency is a solved problem now.