Week 3 of a Claude Code agent running growth: users up from 4 to 37
intensityflow · reddit · 2026-10-11
The author runs a Claude Code agent twice daily to handle growth for a free Chrome extension: it reads the store dashboard, checks comments, finds threads and drafts posts, with nothing published until a human says go and hard caps on posts/replies enforced in code.
Week 3 lessons:
- Overlapping approvals: two sessions both got a go; the morning run posted 10 items, 7 of which duplicated the evening batch. A pre-reply check (reading the thread and the post log) prevented double-posting.
- Password guardrails: Google demanded re-authentication 5 nights straight; the agent can't type passwords, so it correctly reported blocked—costing 5 days of real numbers.
- Stale queue: after 3 days of silence, the agent kept re-drafting replies to cold threads, so the author told it to stop replying to stale ones and start new posts.
Results: weekly users grew from 4 to 37, 52 installs; the guardrails discussion thread drove more store visits than the actual launch post. Open question: how to handle approval queues going stale when the human goes quiet—time-based expiry or something smarter?
More from coding & agent
- Agent swarms race to lower integer multiplication bounds, with kappa breaking 10⁻³ overnight — iskander · 2026-10-11
- Shopping sites are now blocking AI agents Instinct and Muse almost across the board — AccBalanced · 2026-10-11
- GOAT'd Text Generator: call LLMs directly inside ComfyUI to write prompts — ninjaGurung · 2026-10-11
- A 24/7 multi-agent coding setup: Claude Code as co-founder, Codex fleet as builders — EXM7777 · 2026-10-11
- Small business owner seeks agent to auto-reply emails, texts and edit contracts — jimmywalruss · 2026-10-11
- Meta's GitSwarm: Hundreds of Async Agents Sharing a Git Repo Solve All 30 IMOProofBench-Advanced Problems — omarsar0 · 2026-10-11