JS engineer to Lead AI Engineer: Omio's production agentic stack, from MCP rebuild to token costs
js_craft_hq · x · 2026-09-14
A podcast interview with vilvaathibanpb, who went from JavaScript engineer to Lead AI Engineer at Omio in 2024 with zero ML background, shipping the internal AI platform, MCP layer, and consumer agent Omio AI. Key points: dropping LangChain/LangGraph to rebuild on MCP, supervisor vs orchestration agents, bigger models with medium thinking beating mini models on cost, tool-call caps and response summarization, prompt caching at scale, subjective guardrails, loop engineering as a sugar-coated while loop costing 10-20x tokens, RAG today, and why fine-tuning makes sense again by 2027.
More from coding & agent
- Custom harness lets devs mix older models: 'previous models are good enough for daily work' — omarsar0 · 2026-09-14
- Dev open-sources Claude skills repo that turns LLM into a creative partner with a web studio UI — jh3yy · 2026-09-14
- 5-week A/B test: OptMem vs Claude Code native memory — slightly better, rotting differently — _rchaves_ · 2026-09-14
- DAIR.ai Founder: Opus 5 + GPT-5.6 Mix Handles Daily Work, Rarely Needs Frontier Models — omarsar0 · 2026-09-14
- LLM-powered pipeline turns character sheets into cursor-following mascots — jh3yy · 2026-09-14
- Dev argues coding's best form factor is a chat box again since agents manage agents well — xeophon · 2026-09-14