Replit president Michele Catasta predicts the first one-person $1B company by end of 2025 and the end of the prompting era

Michele Catasta thinks the prompting era is nearly over, and Replit is building the product to prove it.

Catasta launched Replit Agent in September 2024, before "vibe coding" existed as a term. The phrase wasn't coined until February 2025. That head start gave Replit room to run through four major agent releases, and Catasta argues the next phase is more significant than the one that just passed.

“I do believe that sooner than later, we're gonna be seeing far fewer apps being used rather than agents... We have several 9-figures companies built on Replit. So it's gonna happen. Yeah. We're almost there... Stop giving it prescriptions, stop giving it rules. Just tell the model what you want to accomplish and then trust the process.”

Post-prompting

The argument is straightforward: the way people have been instructed to use AI — carefully crafting prompts, writing rules, specifying steps — is giving way to goal-setting. Catasta points to Anthropic's own guidance on Claude, where researchers are now telling users to stop giving prescriptions and just state what they want. Replit's product direction follows the same logic. A user sets a high-level goal — say, improve app conversion by 5% — and the agent decides what to change, whether that's copy, images, or pricing layout, then measures until it hits the target. The user never writes a prompt about how to do it.

Catasta calls this the "post-prompting era," and frames it as a shift from apps to agents more broadly. The UIs built over the past few decades were designed around keyboards and mouse clicks because that was the only available input. Agents work more like delegating to a colleague — express intent, set a goal, let the system complete the task end to end.

Deterministic vs. agentic

There's a real cost trade-off underneath all of this. Deterministic software built on Replit — the inventory management tools, internal dashboards, zero-to-one solopreneur apps — remains dominant today partly because frontier model calls are expensive at scale. If hundreds of thousands of employees are hitting an internal tool, routing every interaction through a frontier model is financially unworkable. The case for pre-built, deterministic software isn't just habit; it's token economics.

Catasta expects that balance to shift as models get better at long-horizon tasks. He notes models can now run autonomously for four hours at a stretch, a capability that didn't exist when Replit Agent launched.

The one-person billion-dollar company

Catasta says Replit already has several nine-figure companies built on its platform. His prediction: the first clean, one-person $1B company arrives by the end of 2025.