Devs Automated Mom's Wedding Venue Business With an Open-Source Agent: 120 Hours Saved for $350/Mo
Free_Lead_2704 · reddit · 2026-10-15
Background
The author and a cofounder built an open-source (AGPL) AI coworker; its first user was the author's mom, who runs a wedding venue in southern Italy. Her workflow: receive email inquiries → create leads in a 10-year-old CRM with no APIs → send a custom reply and questionnaire → quote via the CRM → iterate until contract. At 15-20 inquiries/day, each taking 20-30 minutes, this consumed 70-80% of her week.
Technical approach
- API reverse engineering: Instead of hand-written integration scripts, they built a skill + CLI that lets the agent reverse-engineer the APIs of any web app — it worked surprisingly well on the old CRM.
- Automation (not a cron job): Gmail triggers on new email → filters (which the agent set itself by studying what an inquiry looks like) → each step handled by a separate skill.md → every step logged for rollback. Fully agent-driven, no node diagrams.
- Permissioning: To build trust, they added granular, deterministic permissions — per-tool-action control over auto-run vs. approval (e.g. "Gmail – Send Email" requires approval), with push/email approve-deny buttons. After error rates dropped below 2%, the mom granted full access.
- Automation scout: The agent audits run history of all automations; on errors or human corrections it rewrites the system prompt, tools, skills and filters — self-improving.
Results & cost
After 3 weeks: median inquiry processed in 5 minutes; mom only approves quotes for special cases (very large weddings). Runs on gpt 5.6 Luna for $350/mo all-in, saving 120+ hours/month.
Lessons
- Pick one very painful job and truly understand it
- Don't hand-write integrations; let the agent research, write and maintain them itself
- Agent permissions should be extremely granular, down to each action — pick infra accordingly
- Permissioning must be deterministic; telling the agent "don't send the email" in a prompt won't cut it
More from coding & agent
- Claude Code Pushes Urgent Security Update Across All Installations — umeshonai · 2026-10-23
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23