---
title: "Shipping a small product fast with AI coding (Claude Code, Cursor)"
slug: shipping-fast-with-ai-coding
summary: "How solo and non-technical founders shipped in hours to days in 2025–26 — Geosports built by a non-coder with Claude in ~8 hours from a written spec, TrustMRR in under 24 hours with Cursor, Fastlane's stack on Claude Code Max — plus one-feature scope, deadlines (Ship or Die: 155 of ~200 shipped in 30 days) and when to stop adding features."
domain: marketing
tags: [ai-coding, claude-code, mvp, shipping, tech-stack, indie-hacking]
related: [finding-indie-product-ideas, customer-discovery-calls-at-scale, app-mvp-single-problem-scope, app-development-cost-and-tooling-2026, ai-landing-pages-with-claude-code, indie-case-studies-common-patterns]
sources: [yt:s9aLA4IOSZE, yt:m268_-MiUYM, yt:BA_fC95p43c, yt:I-pPhs5Qbsk, yt:n7Obxh9-Rc8]
updated: 2026-10-24
---

# Shipping a small product fast with AI coding

## TL;DR
- Write a spec with the LLM first (research the reference product, list the game/feature rules), then build against it, referring back to it whenever stuck.
- Non-technical? Screenshot every confusing dashboard (Vercel, domains, DNS) and ask the model what you're looking at and whether your input is right.
- Block one uninterrupted 8-hour session to get v1 out.
- Scope to **one headline, one feature** (Marc Lou); complexity makes visitors drop off.
- Public deadlines work: in Marc's Ship or Die cohort, 155 of ~200 people shipped within 30 days.
- After traction, stop adding features and make what exists reliable (Postiz).

## Details
**Geosports — non-coder, a weekend.** Frank (sports creator, not technical) described the game to Claude, had it research MapTap and write a spec, and asked Claude whenever he got stuck; it also recommended tools. ~8 hours on Sunday, then he cleared Monday and shipped.[^1][^2] Stack: Claude Max (20x), Vercel Pro hosting and domain, MapLibre for the globe, Upstash Redis for leaderboards/stats, Google Sheets for the questions, Stripe for Pro subscriptions.[^2] His tips: block off eight uninterrupted hours and lean on screenshots plus questions — the model was right most of the time.[^3][^4] Host Gus's takeaway: AI coding removed the technical blocker, so people who already think about distribution (creators) are best placed to win.[^5]

**TrustMRR — under 24 hours.** Marc Lou dumped his thoughts into Cursor and had AI build a dead-simple site: a "verify your startup" button taking a read-only Stripe API key and a Twitter handle, a per-startup revenue chart and search.[^6] His lesson: one headline, one feature; strip everything else.[^7]

**Fastlane — a small team's 2026 stack.** Convex (backend/DB), Vercel (front end), Railway (content-building microservice), Clerk (auth), Resend (email), Axiom (logs), OpenAI and Claude models for copy and captions, fal.ai for image/video generation, and Claude Code on the 20x Max plan, which Gaurav calls a must for their shipping speed. The non-technical co-founder built an internal analytics dashboard with Claude Code and MCPs.[^8][^9]

**Deadlines and accountability.** Marc Lou's Ship or Die program makes members sign an "oath" to ship a startup in 30 days or be removed without refund. First cohort: ~300 joined, 155 of the first ~200 shipped (75%+), 22,000 commits; 57 connected Stripe and made $33k in 30 days. It later added marketing courses (short-form video, SEO, Facebook ads, taught by friends) with missions verified by uploaded screenshots, and Stripe sponsored $250 credits per member; ~10% get kicked out.[^10][^11][^12][^14]

**When to stop shipping features.** Nevo (Postiz, 30+ social networks) stopped building new features once growth hit, because every feature is a new bug; he wants the app never to break, and suspects agents will learn which tools are reliable.[^13]

## Disagreements / caveats
- Speed-to-ship stories are survivorship-heavy; Ship or Die's stats are Marc's own marketing for a paid program.
- AI-built code still needs security review (the source does not cover it); TrustMRR handles Stripe keys — use read-only keys as Marc did.
- Fastlane spent ~5 months from first V1 to public launch — "ship fast" there meant fast iterations, not a single weekend.

## See also
- [[app-development-cost-and-tooling-2026]] — agency view of AI-assisted builds
- [[ai-landing-pages-with-claude-code]] — the same workflow for landing pages
- [[app-mvp-single-problem-scope]]

[^1]: yt:s9aLA4IOSZE @ 07:13 — Frank Mickelsmith
[^2]: yt:s9aLA4IOSZE @ 08:01 — Frank Mickelsmith
[^3]: yt:s9aLA4IOSZE @ 09:40 — Frank Mickelsmith
[^4]: yt:s9aLA4IOSZE @ 10:27 — Frank Mickelsmith
[^5]: yt:s9aLA4IOSZE @ 15:09 — Gus (Starter Story)
[^6]: yt:m268_-MiUYM @ 00:00 — Marc Lou
[^7]: yt:m268_-MiUYM @ 04:55 — Marc Lou
[^8]: yt:BA_fC95p43c @ 03:58 — Gaurav
[^9]: yt:BA_fC95p43c @ 09:00 — Gaurav
[^10]: yt:I-pPhs5Qbsk @ 00:46 — Marc Lou
[^11]: yt:I-pPhs5Qbsk @ 02:19 — Marc Lou
[^12]: yt:I-pPhs5Qbsk @ 09:32 — Marc Lou
[^14]: yt:I-pPhs5Qbsk @ 11:04 — Marc Lou
[^13]: yt:n7Obxh9-Rc8 @ 09:46 — Nevo David