Short answers
Direct answers to questions founders type into Google, ChatGPT, and Perplexity. Each page starts with the sentence you can cite — then the nuance.
AI products
What is a RAG MVP?
A RAG MVP is one job — “ask my files” — with ingest, chunking, retrieval, citations, and auth so User A cannot retrieve User B’s documents.
Do I need to fine-tune a model for my MVP?
Almost never for an MVP. Start with a hosted model and, if you have documents, RAG. Fine-tune only after you have a real eval set and a job that prompting cannot do.
How do I build a SaaS with AI?
Pick one user job, call a hosted model from a server route, wrap it in Next.js auth and Stripe, and cap tokens. Do not start with agents, fine-tunes, or your own GPUs.
Do I need a chatbot or RAG?
RAG if users ask over their own files. A chatbot if answers live in a short FAQ you own. An agent if the model must call tools (CRM, email, calendar).
Should I use OpenAI or Anthropic for my product?
Pick the model that wins your 20-question eval at acceptable latency and cost. Wrap the API so you can switch. Do not build your company on one vendor’s playground UI.
What is an AI MVP?
An AI MVP is the smallest product that lets real users complete one AI-powered job: accounts, a hosted model, and a usage cap. It is not a ChatGPT wrapper and not a foundation model you trained.
Prototype → production
Can a Lovable app go to production?
Yes, after you export or rebuild into a repo you own and review authentication, row-level security, secrets, and a deploy you can roll back. Lovable itself is not “production.”
Can Cursor-generated code go to production?
Yes. Cursor writes code; it does not ship it. Treat the output like a junior’s first draft: review authz, secrets, migrations, and monitoring before strangers sign up.
What is a production-ready AI app?
A production-ready AI app is software you would let strangers create accounts on: tenant-safe data, server-only secrets, a deploy you can roll back, observed errors, and payments that match entitlements.
Should I use an AI builder or hire an agency?
Use an AI builder to learn the flow. Use a production team (agency, freelancer, or hire) when you need owned code, stores, tenant-safe data, or a deadline. An agency is one path — not the only one.
Is Bolt.new production-ready?
Bolt.new can generate deployable code. “Production-ready” is your auth, data isolation, and ops — not a property of the sandbox. Export early and run the same checklist as any AI-built app.
Web & SaaS
Should I use Next.js or React for an MVP?
React is the UI library. Use Next.js if you have public pages, SEO, or want marketing + app in one repo. A Vite React SPA is fine for a login-only dashboard.
How do I build a web app in 2026?
Write the one user job. Then Next.js, hosted Postgres, real auth, Stripe if you charge, error tracking, and a host you can roll back. No microservices. No AI builder if strangers will pay.
How do I build a SaaS MVP?
A SaaS MVP is one workspace model, auth, Stripe, and one resource table for the job. Next.js + Postgres is enough. Cut admin theatres and “also an AI platform.”
Should I build web or mobile first?
Web first if the job happens at a desk or you need Google. Mobile first if camera, push notifications, or App Store distribution are the product. Doing both in week one doubles the surface.
Cost & time
How much does an AI MVP cost?
Two bills: the product build and inference. A one-workflow chat/API product can start at $3,460 with us. RAG and agents are higher. Token spend must be capped or ads will bankrupt you.
How long does it take to build an MVP?
A tightly scoped product: 15–21 days after kickoff with a team that already knows the stack. Marketplaces, RAG, or regulated domains: 4–8 weeks. Anyone promising Uber-with-drivers in 14 days is selling a demo.
How much does a web app cost to build?
A tightly scoped production web MVP can start at $3,460 with a specialized team. Traditional agencies often quote $30k–$150k for the same loop. No-code is cheaper until the migration.