Jul 15, 2026
Shopify said yes.
Seventh build log. Two days after I wrote about submitting Kōkua and not knowing if it'd get approved, Shopify approved it. It's live on the App Store.
The Spark · three plays every Sunday
You've got more ideas than time — or one you can't tell is real. Sparkfile is how you find the one worth building and kill the rest before they cost you a season, then run it: no team to hire, no app to buy, just you and Claude.
The flagship
Drop in an idea — or three — and Sparkfile runs each one through four honest gates: real demand, a real wedge, a cheap way to validate it, and whether you can actually build and run it yourself. Most ideas don't survive. That's the point: you find the one that does in an afternoon instead of a lost quarter. Free to try, and it runs in your own Claude — nothing to install.
Run your idea through the filter →Then run it
Once you've found the one worth building, the free 90-day roadmap takes you from an empty folder to a running business: three AI cofounders — Growth, Ops, Brand — that you direct as CEO, on Claude alone. A Claude Code path for people who've never opened a terminal, and a chat-only track you can start today. Find it, then run it.
See the 90-day roadmap →The live run
All entries →Jul 15, 2026
Seventh build log. Two days after I wrote about submitting Kōkua and not knowing if it'd get approved, Shopify approved it. It's live on the App Store.
Who it's for
You've got a folder of ideas and no honest way to choose — or you've already picked the thing and you're doing every job yourself: the spec at midnight, the proposal, next week's schedule. Sparkfile is for both ends of that. Find the idea worth building, then run it without hiring a team you can't afford. Written by someone who's done the jobs, not someone selling a future.
The difference
You don't need an orchestration layer, a drawer of third-party agents, or one more SaaS wrapper. The filter runs in your Claude. The team is memory files and prompts you own. Everything here runs on Claude's own tools — Chat, Cowork, Claude Code — because that's already the whole stack. A wrapper would just put a tollbooth between you and the thing that does the work.
From the library
See all →Pull real search-demand signal for your idea, its category, and a competitor — programmatically, no screenshots — and get a plain read on whether interest is growing or dying before you commit a week to building.
Four gates and two human checks that turn 'should I build this' into a verdict (Keep, Bench, or Scrap) before you write a single role file.
Cursor is a full code editor with AI built in; Claude Code is conversation-and-diff-first with no editor to learn. The right pick depends on whether you want to live in an IDE or not.
Skip the benchmark comparisons — the useful test is running your own actual, recurring task on both and judging the output against what you know to be true.
No-code builders are for structured, persistent apps with a defined data model. Cowork is for conversational, judgment-heavy workflows. Most founders need both, for different jobs.
A framework for deciding whether to pay, since exact plans and prices change: usage volume, whether you need Cowork or Claude Code, and how much context you regularly need in one conversation.
The operator
25 years building audio products and companies: Grace Design, HeadRoom, now Warren Labs. These days I ship real products by directing Claude instead of hand-writing every line — and I run every one through this same filter first. The most recent one it greenlit, Kōkua, a Shopify accessibility scanner, is now live on the Shopify App Store underKahakai LLC.
Read the full story →The Spark
Read it in three minutes. Ship something by Friday.