Replit has launched Replit for Business, a product aimed at the companies now discovering that AI agents are easy to start and hard to manage.
The pitch is simple. Employees want to build with agents, IT wants to sleep at night, and someone has to referee.
The launch arrives with backing from a16z, where Marc Andreessen has been arguing that companies should build agent habitats. These are structured environments that keep AI agents from running wild without crushing the initiative of the people using them.
What Replit is actually selling
Replit’s enterprise offering packages orchestration, sandboxing and error recovery into a governed environment, along with identity controls that track which agent is doing what.
Orchestration coordinates multiple agents working on related tasks. Sandboxing isolates them so a mistake in one place doesn’t spill into production systems.
Error recovery is the cleanup crew. When an agent fails, the system is designed to catch it rather than leave a human to discover the wreckage on Monday morning.
Replit’s offering includes SOC-2 compliance, a security audit standard that procurement teams often treat as table stakes.
It also supports Bring Your Own Cloud, or BYOC, across Azure, Google Cloud and AWS. Customers can run Replit’s tooling inside their own cloud accounts instead of handing data to a third party.
The self-driving company
Replit CEO Amjad Masad has framed all of this as part of a broader vision he calls the self-driving company. In this model, AI handles routine operational work so employees can focus on decisions that actually require judgment.
Masad has pointed to examples like automated ticket triage, where agents sort and route incoming requests.
Andreessen’s framing tries to hold two ideas at once. Companies need to contain agent chaos, and they also need to preserve the bottom-up experimentation that makes agents useful in the first place.
The numbers behind the enterprise push
Replit’s business mix suggests this is more than a branding exercise. Enterprise contracts grew to approximately 15% of total revenue in 2026, up from around 2% in 2025.
That is roughly a sevenfold jump in share over a single year. B2B revenue overall has reportedly crossed 50% of the total.
The expansion pattern inside customers is also notable. Companies adopting Replit have reportedly seen their spending double after six months and grow fourfold after one year.
Replit agents are reportedly in use at 85% of Fortune 500 companies.
How Replit got here
Replit was early to the agent wave, launching its first agent, Agent 1, in 2024. Since then, the conversation in enterprise AI has moved from “can agents write code” to “who is responsible when they do.”
The company has also been reshuffling its model partnerships. Replit has shifted toward using OpenAI models as its default provider, moving away from Anthropic.
Replit has also been expanding toward government customers and has appointed a new president of corporate affairs to support that push. The timing lines up with a Gartner forecast that at least 80% of governmental bodies will implement AI agents by 2028.
Government buyers are famously demanding about security and data residency. Features like BYOC and SOC-2 compliance are not optional extras in that market. They are the entry ticket.
What this means for the agent market
The competitive question is who owns that control layer. Replit is positioning itself as the habitat, but cloud giants, model providers and dedicated security vendors all have reasons to want the same role.
Replit’s advantage is that it already has a footprint inside large companies. Its challenge is that the hyperscalers it integrates with through BYOC could decide to build similar governance tools natively.
For investors watching the AI application layer, Replit’s revenue mix is the metric to track. A move from around 2% to approximately 15% of revenue from enterprise contracts in one year shows the direction of travel.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.