Replit CEO Amjad Masad: agents drove 500% metric gains, net dollar retention above 100% on consumer plan
Mar 24, 2025 with Amjad Masad
Key Points
- Replit's V2 agent drove 100% to 500% gains in A/B testing metrics, with above 100% net dollar retention on its consumer plan—retention Masad says is nearly unheard of for consumer subscriptions.
- Replit's stickiness comes from deployment lock-in rather than switching costs, as users stay once projects are live on the platform, distinguishing it from zero-friction coding tools like Cursor and GitHub Copilot.
- Masad argues vertical SaaS is vulnerable to displacement by AI-built tools on Replit, while horizontal platforms like Salesforce with deep integrations remain defensible against AI-generated alternatives.
Summary
Replit CEO Amjad Masad makes a pointed case that the AI coding boom is producing real, sticky revenue — not just inflated ARR numbers that will unwind.
The headline figure: Replit has above 100% net dollar retention even on its consumer plan, which Masad describes as almost unheard of for a consumer subscription. Rolling out its V2 agent has moved key success metrics by 100% to 500% in A/B testing — gains Masad says he has never seen in his career, including his time at Facebook.
Retention vs. the churn fear
Masad is candid that rapid AI revenue growth made him nervous at first. His answer was to shift focus from ARR growth rates to revenue quality metrics. His argument against the VS Code-adjacent tools — Cursor, GitHub Copilot, Windsurf — is direct: switching cost is essentially zero, and users can move between them on the same project in minutes. Replit's stickiness comes from deployment. Once something is live on Replit, users tend to stay.
The beginner guardrails argument
Where vibe coding goes wrong, Masad argues, is when beginners use tools that lack guardrails — pointing to publicised cases of developers losing months of work or leaking API keys. Replit handles git commits automatically and stores API keys in a secure vault. The pitch to non-technical builders is safety by default, not just speed.
Who is actually building on Replit
The Sears Home Services case is the most concrete enterprise example Masad offers. The company, migrating off COBOL, skipped the usual SaaS and low-code modernisation layer entirely and had its operations team build AI tools directly on Replit — including a route-optimisation tool for field workers. Masad frames this as a preview of how firms will be structured: generalist employees who spin up AI agents rather than teams of specialists.
Internally, a couple of Replit engineers built an agent-tracing tool in about a week that has saved significant engineering hours. An HR employee with no coding background built a custom org chart tool with ADP integration and version history in three days.
The SaaS disruption question
Masad sees vertical, niche SaaS as increasingly vulnerable — he says he sees someone on Twitter replacing software that costs $15,000 to $100,000 with a Replit-built tool almost every morning. Ecosystem SaaS with deep integration layers, like Rippling or Salesforce, is harder to replicate. His advice for anyone starting a SaaS business today is to build horizontally and broadly, not narrowly — the Silicon Valley focus dogma works against you when AI can generate vertical tools on demand.
The Vercel dispute
Masad declined to support Next.js in Replit's agent, choosing to back Vite instead, on the grounds that Next.js has drifted toward Vercel optimisation. The existence of the open-source OpenNext project — which exists specifically to make Next.js deployable outside Vercel — and public agreement from Netlify's CEO support his read of community sentiment. He says Vercel overreacted to his comments, and considers the matter closed.
Company structure
Replit now has roughly 65 employees after Masad cut the team in half last year in what he calls founder mode. The company has moved out of San Francisco to Foster City, where it is building out an office campus. The team reduction and relocation are deliberate: Masad wanted to shed executives who were cycling through Silicon Valley hype cycles and build a tighter, more mission-aligned group. The selection filter he describes is blunt — if you think Replit is a toy, you are probably not someone he wants on the team.