Stack picks for solo builders in 2026
A working default stack: framework, hosting, database, auth, payments, email, analytics. What to use, what to skip.
Most stack guides give you ten options for each layer and call it neutral. That's not a guide, it's a Wikipedia article. Below are the picks I'd hand a friend who's starting a new project this week, with my reasoning, knowing I'm allowed to be wrong about a couple of them.
The default stack, layer by layer
Framework: Next.js 15. Runner-up is Astro for content-heavy sites (blogs, docs, marketing), or SvelteKit if you're allergic to React. Why Next over the rest: it's the framework with the deepest hosting story, the most third-party examples, and the most training data baked into Cursor and Claude Code. When you ask an AI to scaffold an auth flow or a server action, it answers in Next dialect. That alone saves hours a week.
Hosting: Vercel for pure frontend or Next.js apps. Railway when you also have a backend service or worker. Vercel is the cleanest path from git push to a public URL, full stop. The moment your project includes a Python script, a long-running queue consumer, or a Discord bot, Railway's single-platform model beats stitching two providers together.
Database: Postgres on Supabase, Neon, or Railway. Pick whichever your hosting platform integrates cleanest with. Supabase if you also want auth and storage in the same dashboard. Neon if you want serverless branching (a fresh DB per preview deploy is genuinely useful). Railway if your services already live there. The pick matters less than you think; the SQL is the same.
Auth: Clerk for the easiest UX, Supabase Auth if your DB is already on Supabase. Clerk's hosted UI components are the closest thing to "auth on autopilot" the ecosystem has. Supabase Auth is a perfectly good second choice if you don't want a second vendor. Never roll your own. Auth bugs are silent and the failure mode is "your users get hacked."
Payments: Stripe. There is no runner-up. Lemon Squeezy if you specifically want merchant-of-record handling for global sales tax (real win for solo founders selling internationally). For everyone else: Stripe.
Transactional email: Resend. Runner-up is Postmark. Resend has the cleanest DX I've seen in this category, React-based templates, and a free tier that covers most apps for months. Postmark wins on reputation and inbox placement if you're sending high volumes from day one.
Marketing email: Beehiiv if it's newsletter-shaped, Mailchimp if it's transactional-list-shaped. Beehiiv if the product looks like "issues sent on a cadence with growth tooling around it." Mailchimp if it looks like "I have a list of customers and I send them announcements sometimes." Don't run both.
Analytics: PostHog. Free tier covers small projects, session replay is built in, feature flags and A/B tests come along for the ride. You will not outgrow it.
Error tracking: Sentry. Free tier is enough until you have real traffic. Set it up before launch, not after.
AI/LLM: Claude or GPT via API. OpenRouter if you want one bill across providers and the ability to swap models without rewriting code.
Editor: Cursor for daily, Claude Code for big refactors. Same combo I'd recommend to anyone shipping a real product.
What not to bother with at this stage
Kubernetes. A custom CI/CD pipeline (Vercel and Railway both have this built in, and theirs is better than yours). A microservices architecture for an app that has zero users. Your own auth. Your own analytics. A real-time engine you don't actually need yet. Most products don't need WebSockets, and the ones that do know it on day one. Adding Pusher or a Socket.io cluster "just in case" buys you a category of bug for free.
Boring is good
The stack above is intentionally unsexy. Every layer has been around for years, has good docs, and has plenty of training data inside Cursor and Claude Code. That last part matters more than people admit: when your AI assistant has seen ten thousand examples of a Next.js + Supabase + Stripe app, it ships working code on the first try. When you pick the trendy alternative on every layer, you debug six unfamiliar things at once when something breaks. Pick boring on the layers that aren't your product. Save weirdness for the actual differentiator.
Escape hatches
There are projects this default stack is wrong for. If you're shipping mobile-native, the conversation moves to React Native or Expo, and Next becomes irrelevant. If you're shipping a desktop app, look at Tauri (lean, Rust core) or Electron (heavy, but every example exists). If you're shipping ML infrastructure (model training, inference at scale, GPU orchestration), this guide doesn't apply at all; that's a different set of choices and a different bill.
The cost reality
This stack at zero users runs you about $0 to $30 a month, mostly because everything has a free tier and the only thing you might pay for is a domain and an LLM API key. At 1,000 paid users, expect $200 to $400 a month total infra across everything above. Compare that to building any one of these layers yourself. Auth alone, done properly, is a quarter of an engineer.
Next up
You have a stack. The next pillar is "Connecting tools without writing a backend," which covers integration patterns: webhooks, Zapier, n8n, and the small amount of glue code that turns these layers into a product instead of a checklist.