The mental model for AI-assisted building has always been sequential: you describe a thing, the agent builds it, you review, repeat. Replit Agent 4 breaks that loop entirely. You can now spin up multiple agents inside a single project, each working on a different deliverable at the same time, and they all share the same context, design system, and backend.

One project, everything you need to ship

The core shift in Agent 4 is what Replit calls Multi-Artifacts: the idea that a project is not a single app but a container for every output a launch requires. Agent 4 lets you create web apps, mobile apps (iOS and Android), landing pages, presentations, and even promotional videos in the same project, all sharing the same context and design system. That last part matters more than it sounds. When you ask the agent to spin up a mobile companion to your web app, you don't need to re-explain the project. While the mobile app was building, you could prompt for a pitch deck and a promotional animated video in a single message, and the agent uses the same project context and design language across everything.

Artifacts are the publishable outputs inside a project: web apps, mobile apps, slide decks, and videos. A project can contain a maximum of 7 artifacts. They share a database, backend, file storage, and deployment lifecycle, which means one artifact writing data is instantly visible to the others, and publishing pushes everything live together.

Parallel agents, not just parallel tasks

The headline feature is true parallelism at the agent level. Instead of one AI working on your project step by step, Agent 4 spins up multiple AI agents that tackle different parts simultaneously, such as authentication, database, front-end design, and back-end logic all at once. When the agents' work overlaps, specialized sub-agents resolve conflicts automatically before merging everything into one project.

There are two flavors of this parallelism: