Replit's Agent 4 Runs Parallel AI Builders Across Every Launch Asset

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:

• Independent parallel tasks: tasks can run in parallel with progress visible and coordinated. Once done, they can be merged into the main project. When changes conflict, Agent 4 uses specialized sub-agents to resolve them.
• Automatic task decomposition: for larger jobs, Agent 4 can split a single task into smaller pieces, work on them simultaneously with sub-agents, and recombine the results, shortening long-running tasks without sacrificing quality.

After kicking off multiple threads, you can sit back and watch the agents work. There's a Trello-style board view where you can see tasks move through three stages: Drafts, Active, and Ready. This gives you oversight without forcing you to babysit each step.

The workflow that changed

Agent 3 had a rigid structure: pick a project type upfront, then watch the AI build it. In Agent 3, the building experience was scoped to apps. Before you could start, you had to pre-select an artifact type, which meant committing to a direction before you'd fully explored the problem. That constraint is gone. Whether you want to create slides, a website, a web app, or even a mobile app, you simply describe what you need and the Agent does the work. You can even build different artifacts in parallel.

The old fork-and-merge collaboration model is gone; teammates now work together in a single shared project with agent-assisted merging. The plan-then-build workflow has evolved into plan-while-building, so you can keep moving forward without waiting for one stage to finish before starting the next. You can also submit requests out of order, and Agent 4 intelligently sequences them and executes in the best order.

What it's actually good for

The clearest use case is the full product launch package. A product launch isn't just an app. It's an app, a landing page, a pitch deck, and a social video. Previously those lived in four different tools with four different contexts. Now they can live in one project.

Practical scenarios where this shines:

• Solo founders building an MVP: for solo founders building an MVP while simultaneously preparing investor materials, the friction reduction of keeping everything in one place is substantial.
• Web-to-mobile expansion: a product built for the web extends to mobile inside the same codebase, no new project, no separate team, no parallel timeline.
• Live data dashboards: Agent 4 connects to external data sources, including Excel and Databricks, so your dashboards pull from live sources.
• External service actions: Agent 4 also connects to external services such as Linear, Notion, and Excel, letting users ask the agent to take action and perform tasks like creating tickets.

The limitations worth knowing

Multi-artifact support has some rough edges. Agent 4 features like the Design Canvas, shared collaboration, and parallel task execution are available in new projects. The ability to create multiple apps in one project is not compatible yet for old projects, but is in process. Old projects still get all the other Agent 4 features.

There's also an important architectural constraint: there is a limit to how many artifacts a project can have. A project can contain a maximum of 7 artifacts. And publishing is all-or-nothing: you cannot publish a single artifact from a project independently. If you need separate deployment cycles, you need separate projects.

On cost, AI-intensive workflows can run up to 40x the subscription cost, and Agent 3's sub-agent model saw some single prompts costing $79. Running multiple parallel agents simultaneously will burn credits faster than single-agent workflows, so it's worth keeping an eye on usage.

What it costs to get in

Parallel task execution is available to Pro and Enterprise users. To mark the launch, Replit made it temporarily accessible to Core users for a limited period. In terms of plans, the Starter plan is free with limited daily Agent credits and lets you publish one app. Core costs $25/month ($20/month billed annually) and includes $25 in monthly credits, 5 collaborators, and unlimited workspaces. Pro costs $100/month with $100/month in credits, 15 collaborators, private deployments, and access to the most powerful AI models. Replit's pricing uses an effort-based model with credits that power AI, deployments, and compute, which gives you flexibility but makes costs harder to predict.

The bigger bet Replit is making

This is Replit betting that code is the foundation of all knowledge work, not just software development. The framing is deliberate: historically, the Agent has focused on building apps. But real work is broader than that. You don't just ship apps, you ship outcomes. By collapsing the gap between the product and everything needed to launch it, Replit is positioning itself less as a coding tool and more as a full creative production environment.

The industry context makes this move legible. Cursor, GitHub Copilot, and Windsurf are all competing hard on the code-editing layer. Replit is climbing up the stack, targeting the moment before a line of code is written and the moment after the last one ships. What separates Agent 4 from Agent 3 is a set of specific, visible improvements: the infinite design canvas, parallel agents that work on different parts of your project simultaneously, multi-artifact creation for building web apps and mobile apps and slide decks within a single project, and a task-based workflow that breaks requests into discrete chunks you can review and approve before anything merges.

The question for teams already deep in existing toolchains is whether the convenience of a single unified context outweighs the loss of best-in-class tools for each individual output. For builders starting fresh, that tradeoff is increasingly easy to make.