How I Ship Fast in 2026: My Modern Next.js Stack

This article tries to explain the stack I use as a baseline for nearly every project, but not for all. It is an opinionated approach that I recommend using in 2026. Additionally, it provides an overview of each part of the stack, and I’ll guide you through its development.

This stack is based on the T3 Stack, which I strongly recommend, since it was my go-to for many projects. However, I’ve developed a customised configuration that reflects the state of the web this year, including some key modern additions that I consider essential.

If you have already read this article and want to start using this stack, you can go directly to this repository.

What is this guide covering

1. Reasons to choose this stack
2. Next.js setup
3. Pagination and layout groups
4. Validation for environmental variables
5. Styles
   1. Shadcn
6. Animations and Transitions
7. Database and ORM
8. Authentication
9. Deployment
10. Wrapping Up

Reasons to choose this stack

The approaches of this stack are AI-first, modern UI, full ownership, and Production-Ready. With this in mind, we are going to be using Next.js and TypeScript, which provide a robust foundation and rapid development. It is worth noting that AI works better with contracts, highlighting the importance of using TS.

Additionally, I have chosen technologies that are well-documented, widely used and IA-friendly. AI agents are more precise when the context is more accessible. So, for example, using TailwindCSS, the AI agent doesn’t have to look for .css files outside the component, since it has inline styles. Another example is Prisma, which requires defining the models in a unique file, so the AI agent has to inspect just one file to get the context of the entire database.

As for the editor I use, I’ve recently started using Cursor instead of Visual Studio Code, because it is built on it, and it helps me with AI agents. However, you can also use Claude Code with VS Code or whatever you prefer.

In 2026, it is not only important to get more confident with your code skills and expand your stack, but also to learn to delegate some tasks to AI and be able to review and correct the results.

So, let’s start setting up our project.

Next.js setup

First, create a Next.js app named modern-nextjs-stack. Then, you can go into this new folder and open it with your code editor.

pnpm create next-app@latest modern-nextjs-stack --yes
cd modern-nextjs-stack

Next.js has already set up most of the necessary technologies.

Routing and layout groups

We are using the Next.js App Router, and here the structure of the folders is important. I recommend that you have the ./app folder only for routing, layouts, groups, api routes, loadings and errors. Other folders have to be outside the ./app folder. An example of this folder’s organisation is the following:

/app              <-- Only for routing, api routes
  ├── api
    ├── route.ts
  ├── auth
    ├── layout.ts <-- Group layout
    ├── page.ts
  ├── (root)
    ├── layout.ts <-- Group layout
    ├── page.ts   <-- Home page
  ├── layout.ts   <-- Main layout
/components       <-- Reusable components
/actions          <-- Server Actions
/lib              <-- Clients of DB (Prisma), utilities, env.ts
/services         <-- Business logic

As you may have noticed, there are many layouts inside the app folder. This approach enables you to have different styles per group of pages. The main layout includes the fonts, global styles and providers. Additionally, I created a group named (root), which provides new styles to the pages within this group, like the Home, About or Contact pages.

Later, I will provide you with a better explanation about the rest of the folders.

Validation for environmental variables

Inside the project, we have to install the following packages:

pnpm install @t3-oss/env-nextjs zod 

The package @t3-oss/env-nextjs is preconfigured for Next.js and allows you to create a client that separates your environment variables by server and client sides. In this case, I’m using zod to validate these variables.

The configuration of this envs client is the next one:

import { createEnv } from "@t3-oss/env-nextjs";
import { z } from "zod";

export const env = createEnv({
  /**
   * Specify your server-side environment variables schema here. This way you can ensure the app
   * isn't built with invalid env vars.
   */
  server: {
    BETTER_AUTH_SECRET:
      process.env.NODE_ENV === "production"
        ? z.string()
        : z.string().optional(),
    BETTER_AUTH_URL: z.string(),
    DATABASE_URL: z.string().url(),
    NODE_ENV: z
      .enum(["development", "test", "production"])
      .default("development"),
  },

  /**
   * Specify your client-side environment variables schema here. This way you can ensure the app
   * isn't built with invalid env vars. To expose them to the client, prefix them with
   * `NEXT_PUBLIC_`.
   */
  client: {
    // NEXT_PUBLIC_CLIENTVAR: z.string(),
  },

  /**
   * You can't destruct `process.env` as a regular object in the Next.js edge runtimes (e.g.
   * middlewares) or client-side so we need to destruct manually.
   */
  runtimeEnv: {
    BETTER_AUTH_SECRET: process.env.BETTER_AUTH_SECRET,
    BETTER_AUTH_URL: process.env.BETTER_AUTH_URL,
    DATABASE_URL: process.env.DATABASE_URL,
    NODE_ENV: process.env.NODE_ENV,
  },
  /**
   * Run `build` or `dev` with `SKIP_ENV_VALIDATION` to skip env validation. This is especially
   * useful for Docker builds.
   */
  skipValidation: !!process.env.SKIP_ENV_VALIDATION,
  /**
   * Makes it so that empty strings are treated as undefined. `SOME_VAR: z.string()` and
   * `SOME_VAR=''` will throw an error.
   */
  emptyStringAsUndefined: true,
});

Styles

In this project, we are using TailwindCSS as our utility-first CSS framework. As you know, the order of utilities is relevant; it is necessary that you first install Esben Petersen’s Prettier - Code formatter extension to VS Code and set it up as your formatter by default. This setup is done with the following settings.json file:

{
  "editor.defaultFormatter": "esbenp.prettier-vscode",
  "editor.formatOnSave": true,
  "editor.codeActionsOnSave": {
    "source.fixAll.eslint": "explicit"
  },
  "prettier.documentSelectors": ["**/*.{js,jsx,ts,tsx,html,css,scss}"]
}

Then, we have to add Prettier and the plugin for TailwindCSS to this project:

pnpm add prettier prettier-plugin-tailwindcss

Now, we only need to add Prettier’s configuration:

/** @type {import('prettier').Config & import('prettier-plugin-tailwindcss').PluginOptions} */
export default {
  plugins: ["prettier-plugin-tailwindcss"],
};

Shadcn

Shadcn is not only open source, but also open code. I see Shadcn as a modern set of rules and a baseline to build your own component library.

Under the hood, Shadcn uses a primitives library of your choice, like Radix UI, to handle the heavy lifting. By providing unstyled, accessible foundations along with their core functionality, allowing Shadcn to focus on the design layer.

Every time you need a component, Shadcn hands you a great-designed component out of the box. These components are open for modification, so if you need to change their behaviour or style, you only have to override or extend the component code.

Last but not least, the principles on which Shadcn is based enable AI to assist you in editing or creating components, because AI has access to the component code, but also because they are composable, predictable and have a common style.

To start using Shadcn, you need to run the next command within this project:

pnpm dlx shadcn@latest init --template next

You will need to choose a component library and a preset of your preference. In my case, I selected Radix and Nova, respectively.

The final step of our styling is to set up the fonts correctly. Next.js has already configured them into the root layout. However, we need to fix the name variable to the one used in the global.css file, for example, like this:

const geistSans = Geist({
  variable: "--font-sans",
  subsets: ["latin"],
});

const geistMono = Geist_Mono({
  variable: "--font-geist-mono",
  subsets: ["latin"],
});

Animations and Transitions

Even if you don’t know how to use animations, I encourage you to install 2 packages for this project. In this way, you can specify the packages you want to use when you ask an AI agent to animate your components.

The first one is Motion (previously Framer Motion), which hands you a simple-to-use animations API. A non-animated website might seem boring, so I recommend you use some animations.

The second one is up to you. Next Transition Router has a simple API to create animated transitions between pages.

Finally, I would like to recommend Next Themes if you want to provide a dark mode on your website.

Database and ORM

To this stack I’ll be using PostgreSQL. This project includes a file ./start-database.sh , which runs a PostgreSQL database within a Docker container. You must set up the next env variable:

DATABASE_URL="postgresql://postgres:password@localhost:5432/modern-nextjs-stack"

Once you have your database running, you can set up your favourite ORM. In this case, I will be using Prisma. To get started with Prisma, you have to install the following dependencies:

pnpm add prisma @types/pg --save-dev

pnpm add @prisma/client @prisma/adapter-pg dotenv pg

Then we can initialise Prisma in the project:

pnpm dlx prisma init

This last command generates the prisma.config.ts file and the ./prisma folder. This last folder contains the schema.prisma file, where you have to define the models of the database. Now, we can generate the Prisma client:

pnpm dlx prisma generate

This command will generate the folder ./lib/generated/prisma. It contains both the Prisma client and the models, types, etc. So, you need to run this command each time you make changes to your database, especially when running migrations.

Now, it is time to configure this Prisma client for the project:

import "dotenv/config";
import { PrismaClient } from "./generated/prisma/client";
import { PrismaPg } from "@prisma/adapter-pg";

export const pgAdapter = new PrismaPg({
  connectionString: process.env.DATABASE_URL,
});

const createPrismaClient = () =>
  new PrismaClient({
    adapter: pgAdapter,
    log:
      process.env.NODE_ENV === "development"
        ? ["query", "error", "warn"]
        : ["error"],
  });

const globalForPrisma = globalThis as unknown as {
  prisma: ReturnType<typeof createPrismaClient> | undefined;
};

export const db = globalForPrisma.prisma ?? createPrismaClient();

if (process.env.NODE_ENV !== "production") globalForPrisma.prisma = db;

From now on, you can add your database models to the schema.prisma. Then, you only need to run a migration and generate the Prisma client, like this:

pnpm dlx prisma migrate dev --name name-of-the-migration
pnpm dlx prisma generate

It is worth noting that Prisma is already connected to your database running on Docker.

Authentication

To manage the authentication, I prefer to use Better-Auth. This library is intrinsically type-safe, includes a comprehensive set of features out of the box and offers plugins that save you days of development. Moreover, it runs on your own infrastructure, and you have full control over the auth system.

Don’t worry about setup or maintenance because it is very simple. Better-Auth integrates incredibly with Prisma, and its CLI can generate the basic models you need for authentication directly on your Prisma schema. Furthermore, if you need a dashboard to manage your users in the future, you can have one in the Better-Auth infrastructure.

Let’s start by adding Better-Auth to the project:

pnpm add better-auth

Now, you need to add the following environmental variables to your .env file:

# Better Auth
# Secret used by Better Auth. You can generate a new secret using the following command:
# openssl rand -base64 32
BETTER_AUTH_SECRET=""
# Base URL of your app
BETTER_AUTH_URL="http://localhost:3000"

Let’s configure Better-Auth, creating the following file:

import { betterAuth } from "better-auth";
import { prismaAdapter } from "better-auth/adapters/prisma";
import { db } from "./db";

export const auth = betterAuth({
  baseURL: {
    allowedHosts: ["localhost:3000", "*.vercel.app"],
    protocol: process.env.NODE_ENV === "development" ? "http" : "https",
  },
  database: prismaAdapter(db, {
    provider: "postgresql", // or "sqlite" or "mysql"
  }),
  emailAndPassword: {
    enabled: true,
  },
});

export type Session = typeof auth.$Infer.Session;

Additionally, create an auth client:

import { createAuthClient } from "better-auth/react";

export const authClient = createAuthClient();

export type Session = typeof authClient.$Infer.Session;

So, on the client side, you can use the authClient directly. On the server side, I would recommend creating the next function, which includes the headers:

import { auth } from "./auth";
import { headers } from "next/headers";

export const getSession = async () =>
  auth.api.getSession({ headers: await headers() });

Better Auth requires an API endpoint to handle authentication requests like sign-in, sign-up, and sign-out. This is why we need to create the next file:

import { auth } from "@/lib/auth";
import { toNextJsHandler } from "better-auth/next-js";
export const { POST, GET } = toNextJsHandler(auth);

After this configuration, it is necessary to add the authentication models (User, Session, Account, and Verification) to your schema.prisma file. Let’s do it using Better-Auth’s CLI:

pnpm dlx auth generate

Finally, we only need to generate the first migration of the database:

pnpm dlx prisma migrate dev --name init
pnpm dlx prisma generate

At this point, the related tables are created in your database.

This project will include examples for the pages: sign-up and sign-in. So, I suggest reviewing the project repository.

Deployment

There are many options to deploy your web app. One of my favourites is Vercel, because you only have to connect your GitHub account, select the correct repository and add the environment variables for production. Then, Vercel will automatically manage the CI/CD, so you don’t have to think about it.

However, there is one more important thing to solve. When you generate a new migration locally, the change will not be directly reflected to your production database. You must create, for example, a Github action for it. This is why I added a Github workflow, which will run each time you push a migration to your main branch. This workflow will not run until you configure the secret DATABASE_URL with the production database URL to the github repository.

name: Prisma Migrate on Push

on:
  push:
    paths:
      - prisma/migrations/**
    branches:
      - main

jobs:
  migrate:
    if: "${{ secrets.DATABASE_URL != '' }}"
    runs-on: ubuntu-latest

    steps:
      - name: Checkout repo
        uses: actions/checkout@v4

      - name: Setup Node
        uses: actions/setup-node@v4
        with:
          node-version: 20

      - name: Install pnpm
        run: npm install -g pnpm

      - name: Install dependencies
        run: pnpm install --frozen-lockfile

      - name: Apply Prisma migrations
        run: pnpm dlx prisma migrate deploy
        env:
          DATABASE_URL: ${{ secrets.DATABASE_URL }}
          NODE_ENV: production

Wrapping Up

This year has shown us that we have new team members, AI Agents. We need to evaluate not only the team’s capabilities to decide the technologies and architecture, but also we have to understand how an AI Agent works better to take more advantage of it.

This project tries to collect the best practices and integrate technologies that are more relevant in 2026. Remember, it is an opinionated stack, so every step depends on your requirements or use cases. For example, I am a huge fan of TRPC. However, I didn’t consider it to shape this stack because it is helpful when requests are made from both the client side and server side. Currently, it is possible to create a complete web app using only server actions, so you decide to integrate it.

In an upcoming blog post, I will create a web app using this stack, leveraging server actions that can drastically boost the performance of your apps.

This stack is available as a template that you can use directly from my GitHub repository. See the image to find the button “Use this template“.