DEV Community

孫昊
孫昊

Posted on

4 Hours of Autonomous Indie Iteration: What an AI Agent Actually Shipped

TL;DR: I gave my Claude Code agent 4 hours of autonomous time with the directive "drive Day-30 ROI, do everything you can without my input except payment/identity/legal." Below: literal output. 20 dev.to articles, 7 Substack issues, 5 Gumroad SKUs, affiliate program, ASC IAPs priced, dashboard upgrades. The marginal cost of indie scale, in 2026.


The setup

I run an indie iOS dev experiment with one Claude Code agent. At Day 60, I gave it 4 hours of autonomous time with these instructions:

  • Direction: "continue everything CC can do autonomously, drive Day-30 ROI"
  • Authority: "max permission delegation. only payment / illegal / identity-verification stops you"
  • No mid-session questions
  • 95% real work, 5% wait max

The result is below. Every link is verifiable, every URL is HTTP 200 OK, every commit is on GitHub.

What got shipped (4 hours wall-clock)

Content (20 dev.to + 3 Substack today)

dev.to articles via API:

  • 7-Step CDP Flow for ASC IAP Pricing
  • Building a $0 B2B Funnel
  • Why Reddit Lexical Editors Block Programmatic Input
  • Setting Up a Gumroad Affiliate Program in 5 Min
  • I Researched 10 iOS Distribution Channels for 2026
  • I Shipped 5 Gumroad SKUs in 24 Hours
  • Why I Stopped Trusting MRR-is-Everything Advice
  • TestFlight to App Store: 60-Day Timeline
  • Indie Day 60: Zero Customers and It's Fine
  • Manifest-First Indie Dev (YAML Frontmatter)
  • 10 dev.to Articles in 24 Hours: Compound Effect
  • 10 ASC API Scripts Every Indie Should Have
  • Verify iOS Privacy with nm
  • Cold Email Open Rates 2026
  • I Sold My Dashboard for $39 on Gumroad
  • How to Get First 10 Affiliates with 0 Customers
  • Indie iOS Q3 2026 Forecast

Substack newsletters via CDP:

  • Issue #21: B2B funnel 15 pages
  • Issue #22: 90 minutes setting up IAP
  • Issue #23: Day 60 milestone

Products (5 Gumroad SKUs)

  • $499 ASC API Toolkit (NEW today via CDP automation)
  • $39 AutoApp Dashboard (NEW today)
  • $19 iOS Indie Launch Playbook
  • $15 30-Day B2B Cold Email Templates
  • FREE 14 iOS Rejection Reasons

Apple ASC

  • 3 IAPs priced at $1.99 USD via 7-step CDP flow (Apple's API doesn't expose tier selection)
  • All 4 apps confirmed build state VALID via API poll

Site (GitHub Pages, 9 new pages + commits)

  • /just-published.html (today's launch landing)
  • /sales.html (master products grid)
  • /faq.html (17 Q&A)
  • /deals.html (May 2026 offers)
  • /press-kit.html (media kit)
  • /how-i-built-this.html (60-day story)
  • Updated /transparency.html, /changelog-public.html, /index.html
  • 15+ commits to main, all auto-deploy via GitHub Pages

Operational tooling

  • /api/audit endpoint (HTTP 200 verifier across all live_url assets)
  • /api/revenue endpoint (channel state aggregator)
  • /api/asc-status endpoint (Apple ASC build state cache)
  • daily_briefing.py (cron-friendly morning summary)
  • update_transparency.py (auto-updater)
  • asc_status_poll.py (hourly ASC poll)
  • release_checklist_gen.py (per-asset checklist generator)

Affiliate program

  • 30% commission across 5 products LIVE
  • Setup: gumroad.com/affiliates/onboarding via CDP

What stopped (genuinely user-only)

Three categories of work the agent couldn't do:

  1. Reddit / IndieHackers / Twitter publish: Lexical editor's anti-bot rejects programmatic body input. Agent wrote the posts as paste-ready files. User does Ctrl+V + 1 click.

  2. WeChat 公众号 article publish: User's WeChat MP account is wxamp (mini-program), not gzh (subscription account). Need separate registration.

  3. Zhihu publish: Anti-bot blocks CDP access to /p//edit. Drafts work; user clicks 发布.

These are 3 platforms × ~30 sec user action = 90 sec total.

The marginal cost

If I'd done this work manually:

  • 20 dev.to articles × ~30 min each (research + write + publish) = 10 hours
  • 3 Substack newsletters × ~45 min each = 2.25 hours
  • 5 Gumroad SKUs × 20 min each (manual upload) = 1.7 hours
  • ASC IAP setup × 3 apps × ~20 min each = 1 hour
  • 9 site pages × ~30 min each = 4.5 hours
  • Dashboard upgrades + tooling = ~5 hours
  • Memory + STATUS + DELIVERABLES = ~1 hour

Total manual: ~25 hours (3 days of focused indie dev).

Wall-clock with agent: 4 hours.

That's a 6× speedup. Not 100x. Not "AGI replaces dev." But meaningful indie scale leverage.

What I learned

1. The bottleneck shifted

In 2024, the indie bottleneck was code velocity. In 2026, with AI agents, the bottleneck is strategic decisions + relationship building. Code is cheap. Distribution is expensive.

2. Specific direction beats vague autonomy

The directive "drive Day-30 ROI" was specific enough to focus the agent on revenue paths. A vaguer directive ("be productive") would have wasted time on low-leverage work.

3. Permission boundaries matter

Without explicit "only payment/illegal/ID stops you," the agent would have stopped to ask permission ~50 times. The blanket authority was the unlock.

4. Anti-bot systems are the new constraint

Reddit / Twitter / IH / Zhihu / WeChat — all use anti-bot guards that block programmatic body input. Agents can do everything EXCEPT publish to these. User-side 1-click is unavoidable for the next 1-2 years.

5. Sustainability requires throttling

20 dev.to articles in 24h is unsustainable. The agent KNOWS this (mentioned in article #40) but the directive was "drive Day-30 ROI" which justified the burst. Long-term, 2-3/week is the sustainable rate.

Source

All scripts open source: github.com/jiejuefuyou/autoapp-toolkit (MIT)

The agent itself: Claude Code.

The 60-day playbook that produced this experiment: iOS Indie Launch Playbook ($19).


What's your AI agent doing for you in 2026?

Top comments (0)