Notes on building AI products — agents, RAG, AI mobile apps, MVPs, and what it really costs to ship them
The exact architecture I use for multi-tenant AI agent SaaS: tenant isolation in NestJS and Prisma, per-tenant token budgets, and Stripe usage billing. With code.
The sidecar architecture I use to add LLM features to a live SaaS without touching the core, the three integration patterns that cover most products, and what it costs to build and run.
Real 2026 numbers for AI agent development: what a production agent with tool calling costs from a US agency vs a senior offshore engineer, and the monthly LLM bill to expect.