The Curious Programmer

Lately, I’ve been thinking a lot about how weird it feels to ship software the same way we did five years ago—especially when the tools we use today can practically read our minds.

AI is here. Not in the “sci-fi overlords” sense. In the “if you’re still stitching boilerplate together by hand, you’re wasting hours you’ll never get back” sense.

So if your stack doesn’t natively speak AI, you’re dragging a parachute behind your product velocity. Time to cut it loose.

When I evaluate tools now, I’m looking for three things:

• Zero drag – No yak shaving. I want to build.
• Sharp defaults – Give me convention over configuration.
• AI-native – Not tacked-on, not duct-taped. First-class primitives.

This is the stack I use to go from idea to real-world feedback in hours—not weeks.

---

🧠 Front-End

Next.js 15 (React 19)

• React Server Components + App Router = buttery SSR/ISR.
• Pages, APIs, edge functions—all in one mental model.
• Turbopack dev cycle = basically warp speed.

Tailwind CSS + shadcn/ui

• Design freedom without chaos.
• Accessible, customizable components built on Radix—production-ready out of the box.

Vercel AI SDK

• Drop-in <Chat /> components, streaming responses, state handling—done.
• Real-time AI UIs without rewriting half your app.

Electron (Optional)

• Want a desktop version of your AI app? You’re already halfway there.
• Works great for local model inference, offline access, or native OS integrations.

---

🧠 API Layer

tRPC v12 or Next Server Actions

• End-to-end types: from form inputs to DB writes.
• RPC model removes 90% of client/server boilerplate.

LangChain.js + LangGraph

• Tool use, multi-agent logic, memory, and RAG built-in.
• Like Redux and React Router for AI workflows.

Vercel AI Gateway

• One endpoint to rule all LLM providers.
• Centralized billing, caching, throttling, failover—all managed.

---

🧠 Persistence

Supabase Postgres + pgvector

• Relational meets vector search.
• Real-time updates, built-in storage, full SQL.

Drizzle ORM

• Schema-first, strongly typed migrations.
• You write code once and it reads like documentation.

---

🧠 Auth & Security

Auth.js (NextAuth v5)

• Full OAuth support, typed sessions, edge-optimized.
• Plays nice with Supabase RLS for fine-grained control.

---

🧠 DevOps & Observability

Vercel

• Zero-config deploys. Preview URLs per push.

Bun + Turborepo

• Lightning-fast local builds, monorepo-friendly.

Sentry + LangSmith + OpenTelemetry

• Know when stuff breaks, trace AI behavior end-to-end.

Modal or Replicate

• GPU-powered model hosting without running your own infra.

---

Why This Stack Works

• 🧠 TypeScript Everywhere – One language. Zero context switching.
• 🌍 Serverless-first – Focus on product, not provisioning.
• 🤖 AI-native from the jump – Streaming, vector search, agent logic baked in.
• 📦 Rich OSS ecosystem – You’re never stuck on an island.
• 🔁 One codebase, many surfaces – Web today, desktop tomorrow, mobile when you feel like it.

---

Try It in 5 Seconds

Spin up a fully-wired version of this stack with my CLI tool:

👉 create-go-fast-app on GitHub
👉 npm package

npm create go-fast-app@latest

Custom AI stack, no config headache.

⸻

This is the stack that lets you build at the speed of thought—and deploy at the speed of AI.

Got a stack you think beats it? I’d love to see it. Bonus points if it doesn’t involve YAML.