Dev shares how he built AI agents that run marketing campaigns end to end
Few_Benefit_4853 · reddit · 2026-10-03
A developer built a platform where AI agents handle marketing campaigns end to end for agencies, and shared the architecture:
- RAG onboarding under 90 seconds: KPIs, brand voice and competitor info feed a shared knowledge base that every agent pulls from, keeping copy on-brand.
- One agent per job: separate agents for ad copy, scheduling, spend, SEO and reporting — small agents are far easier to debug than one giant prompt.
- Logging builds trust: every step is logged so owners can see what ran and why, which is what makes clients comfortable letting agents act autonomously.
Stack: Next.js, FastAPI, LangChain, PostgreSQL, AWS, Stripe. Biggest lesson: split work into small agents, give them one shared source of context, and keep everything visible to a human.
More from coding & agent
- Copilot Home is a nested if, Autopilot is while(true): one analogy explains Microsoft's AI lineup — pswider · 2026-10-03
- Production Agentic RAG course hits 9,275 stars on GitHub — jamwithai · 2026-10-03
- Cloudflare open-sources agent workspace cloudflare-os with 10,387 stars — cloudflare · 2026-10-03
- Indie Fishing Game Built Entirely With Blender MCP and Claude, Playable Demo Coming — sidahuj · 2026-10-03
- 52.6k-star GitHub project adds marketing skills to Claude Code and AI agents — lxfater · 2026-10-03
- Anthropic Academy releases free courses on Claude Code subagents and MCP — anthara_ai · 2026-10-03