Replit CEO Amjad Masad on enterprise growth, open model routing, and a new multimodal design product

Replit CEO Amjad Masad on design, model routing, and enterprise momentum

Replit is pitching itself as the platform where vibe coding graduates from consumer excitement to enterprise workflow automation. Amjad Masad argues the shift is already happening, and the clearest evidence is that enterprise has become Replit's fastest-growing segment.

Replit Design and the post-prompting thesis

The newest product is Replit Design, a multimodal design tool Masad describes as the first in the market to let users combine the best models for different design tasks. Some models perform better on SVG, others on animation or CSS, so Replit evals across all of them and routes accordingly. Users can also override and choose a specific model, including Kimi K3, which Masad singles out as the first model he has seen that genuinely reasons about design — thinking through page structure, layout, and aesthetics rather than just generating output.

The interface is built around what Replit calls ambient intelligence, with Movin integration providing a free mood-board tool. Users drag and drop inspiration from top UI/UX sites onto a canvas, then point and click to generate variations, additional pages, or different aesthetic directions. Masad's framing is that most people have taste but lack the language to express it — prompting, like coding before it, is still too low-level relative to how people actually think. The direction he is pushing is a post-prompting model where users specify intent and agents handle the rest.

“Masad: 'For two years, Anthropic was the undisputed king of coding agents. Right now, it's not just GPT that's caught up, but also open source models — Kimi k3 is an amazing model, sometimes even better, at maybe a quarter of the price.' On enterprise: 'Vibe coding has been proven out, and now we've gone from consumer excitement last year to real adoption in enterprise — people are saying this is the most impactful tool, I used to do this for twenty hours a week, now I built an app and do it for zero hours.'”

Model routing and enterprise control

Enterprise customers are asking Replit for granular model control, and Masad says the demand is driven partly by frustration with big AI labs locking them into large contracts and what he describes as a "token maxing" dynamic where individual employees run up significant spend with no guardrails. He cites examples of single employees costing companies between $10,000 and $20,000 a month.

The Replit pitch is outcome-focused routing with spend controls and task-specific guardrails. In design mode the agent stays in design mode; in app-building mode it stays scoped to that task. Masad says companies are now moving away from giving employees blanket access to Claude and replacing it with Replit precisely because legal and finance teams were burning tokens with no structured outcome. Enterprises are asking for model selection on a per-user and per-group basis, so designers get Kimi K3, front-end engineers get a different default.

The enterprise ROI argument

The competitive framing Masad uses against tools like Microsoft Copilot is direct. He says Copilot saves ten or fifteen minutes here and there; Replit automates entire workflows. His example is an employee who spent twenty hours a week on a process, built an app in Replit, and now spends zero hours on it. That's the story he is taking to enterprise buyers, and he says it's resonating.

Replit's stack depth — hosting, deployment, security agents, admin controls — is what makes the enterprise pitch credible. It is not a single-function AI wrapper; it is a full environment, which matters for IT and procurement conversations.

Game development and physical design

Masad acknowledges Replit built a game development stack about eighteen months ago that didn't land well because models were poor at 3D reasoning. He believes Kimi K3 and Claude Opus 4 are starting to change that, and says a large community is already doing 3D and game work on Replit. A recent CAD integration now lets users design physical products, and Masad says some have already gone from a Replit design to an actual manufactured object. He sees game development, robotics, and physical industrial design as adjacent opportunities.

On AI and economic empowerment

Masad is direct that the dominant public narrative around AI job loss is distorted. He argues there is a perverse incentive for some players to generate negative headlines because fear creates demand for regulation, and regulatory capture benefits incumbents. His counter-narrative is the person who used Replit to build products for business owners in real time and climbed out of $50,000 in debt. Masad says he himself — someone who has never trained a machine learning model — is now training models with Replit, including a chess LLM he claims is already outperforming GPT-class models on chess.

Takeaway: Replit's growth story is shifting from consumer viral moments to enterprise workflow replacement, with model routing and spend control emerging as the commercial wedge. The Kimi K3-powered design product is the most concrete new product signal, and the broader bet is that ambient, intent-based interfaces displace prompting the same way prompting displaced raw coding.