The $9B startup that wants to create a billion new developers
In a Nutshell
Replit, a $9B no-code app builder, enables anyone literate to create deployed, scalable software via "vibe coding" with AI agents (latest: Agent 4), evolving from in-browser IDEs to multimodal natural language interfaces that abstract code entirely. Users—domain experts, creators, and enterprises—build personal apps, vertical SaaS (e.g., health tech, pool maintenance), and internal tools (e.g., CPQ systems saving millions), driving 7% WoW growth and replacing outdated setups like MS-DOS or offshore dev. Future: post-prompting world with autonomous agents every 6 months, fostering "builder" generalists who deputize AI for explosive productivity in a company of founders.
These notes were generated by AI and may contain inaccuracies.
Amjad Masad, CEO and co-founder of Replit, joins the discussion. Replit is the leading no-code app builder for consumers and enterprise. They raised a $400 million Series D at a $9 billion valuation.
Replit's ambition: anyone who can read and write can turn an idea into a deployed, hosted, scalable app without worrying about technical aspects. Mission pursued for 10 years.
- Initially solved development environment setup.
- Then solved deployment environment.
- In September 2024, Replit became the first vibe coding product, abstracting away code entirely with a coding agent interfaced via natural language.
- With Agent 4, added design interactions: write, comment, drag-and-drop on a canvas.
Exploring multimodal interactions (video, audio) to express ideas that turn into real, secure, scalable software—not toy software.
Replit bridges dev tools and consumer apps, first not marketed toward engineers.
Amjad started coding young, built first business at 13-14. Developer tools got worse over time—from simple BASIC to nightmare web app setups post-college.
Built tools focused on joyful creation over accidental complexity:
- First in-browser IDE (initially open source).
- Worked on React and React Native.
Goal: make programming accessible, updated to create a billion new developers.
Developers often like the pain (craftsman building tools). Primary users:
- Tech-adjacent: product managers who coded years ago.
- Designers blocked by engineers.
- Entrepreneurs with ideas but hamstrung by technical needs.
In 2023, explicit goal: not targeting developers, building for creators—AI-native developers creating without managing every system component.
Reminiscent of VB6, which enabled non-coders. Programming got worse (VB6 better than React/Webpack setup). Goal: make programming great again.
Categories: personal software, enterprise software, entrepreneurial products.
Examples:
- Physical therapist and husband built sophisticated app for fascia release: body scans, range of motion tracking on 3D body model. Spent hundreds of thousands offshoring, then built on Replit—best health tech app seen.
- Pool maintenance SaaS by founder from pool family business.
- Sports clubs software replacing MS-DOS apps.
Domain experts closest to problems now build solutions, improving economy, wealth creation, productivity.
Personal software:
- Mom in Korea built app for child's rare condition.
- Family healthcare, wearable data tracking, physical activity.
- Chore hero iPad app ranking kids' chores.
Enterprise:
- Product development: faster iteration under AI pressure. Whoop increased ideas tried by order of magnitude (from 100 ideas/5 tried to 50 tried).
- Internal tools/line-of-business apps: sales automations, revops handling CRM/Gong/data flows, quote configurators, saving hundreds of thousands to millions vs. SaaS silos.
Shift like developers with Stripe, PagerDuty: now product managers, designers, ops empowered to bring software.
- Consumer/PLG: People build personal apps on weekends, realize code solves problems—neurological shift. PLG with referrals.
- Sales-assisted: Champions bring to work; help with hackathons, boss convincing, AI education/evangelism.
- Enterprise top-down: Evaluating vibe coding tools; Replit wins on security/compliance trust.
Can build SaaS, consumer, automation products comfortably without technical knowledge.
Not focused on new cloud platforms or ML systems (though versatile VM/general agent allows sophisticated builds with tech knowledge).
Success stories: consumer apps, vertical SaaS, Replit-native agencies (e.g., Iceland agency 60-70% cheaper/more effective). Internal tools: org charts, CPQ systems integrating HubSpot/Salesforce.
Integrations: Partner-built, plus skills/MCP revolution—agent searches vetted skills (e.g., Stripe), downloads like Neo in Matrix.
Different from dev tools: need to show/educate what's possible.
- DevRel as educators.
- Simple docs, heavy content/video (like Stripe).
- Agent as brainstorming partner for non-devs.
- Canvas/visual interfaces, variation buttons.
Enterprise: Free hackathons with AI-excited groups to create champions.
Ideal champions: entrepreneurial, influential internally, founder mindset (resourceful, integrates tools)—sells across sales/marketing/product/design.
YC realization: achieve massively in 3 months hyperfocus (Sam Altman's advice).
Entered YC as command-line Replit; exited with web dev, hosting, IntelliSense, IDE features—whiteboard countdown to Demo Day.
Now: agent releases every 4 weeks with all-hands intensity.
7% week-over-week growth for new products.
Life-changing:
- Pre-YC: $500K raised, VCs avoided.
- Post-YC: Demo Day intros; rejected 3-4 times, then invited after Hacker News (direct Paul/Sam relationship).
- Asked for a16z intro, led seed after breakfast pitch.
- Expanded network, gave company life—might have quit without.
Agent 4: Fourth major revision.
Predict AI step changes twice a year:
- 2025: Vibe coding early, autonomous late (Opus 4.6, OpenClaw).
- 2024: Mid—Claude, Claude Sonnet enabled code gen (vs. GPT-3.5 laziness); late—long horizon reasoning.
Align roadmap: every 6 months new agent version, predicting/pushing edges.
Agent 3: Most autonomous—runs 2-4 hours; big prompt, leave, return to fully built software. Updated platform accordingly.
They rewrote everything on the backend to support long-running containers in the background doing work while the user is not in front of the computer. Although true autonomy didn't arrive until November and December, Replit Agent demonstrated where the world is headed by September.
With Agent 4, they introduced parallel agents, making it potentially possible. The downside of autonomy was putting in a big prompt and just watching it work, unsure what to do next. The solution: design and kick off other work, chat with the agent, and plan asynchronously. They built a multi-agent architecture, solving merge conflicts and other challenges.
Asynchronous Design Agent and Canvas Interface
As the agent builds, it unblocks the user from designing. They created an asynchronous design agent with a canvas interface to explore the next page or feature while the agent works in the background. Once ready, kick it off into another thread. This creates a state of flow, as agents are slow but work asynchronously.
Solving parallel agents also solved teamwork: fork an entirely new VM for each person jumping into the session, allowing parallel work. The orchestrator subdivides tasks, enabling prompting in the same chat window. With the canvas, seeing other cursors makes the product feel live.
When making a mobile app, website, deck, or video, all should have project context. Previously, different tools were needed for each. Now in Replit, after building a web app, say "make a mobile app"—it generates and lays it out on the canvas. Deploy deploys the web app to the web, mobile to TestFlight or Android. Now run your entire company on Replit.
We're headed to a post-prompting world. People give high-level goals like "optimize my marketing funnel." Prompting remains a skill layer for interactive work, but high-level commands will dominate—e.g., Agent 5 or sooner: "Every day build me a SaaS company, try to market it, make revenue."
- Understanding what is possible: Play with tools, adopt a playful mindset, stay plugged in.
- Not giving up: Retry in a month what fails today; Replit users should try again in a couple weeks.
- Idea generation: Be generative, think of problems to solve, stay creative. Small-time entrepreneurs like Pieter Levels build timely products for millions, then iterate as relevance fades.
Made many mistakes. Culture is vital—they screwed it up, reset with layoffs. Be honest about product-market fit: Getting users or revenue is amazing, but true PMF is explosive. Avoid delusion; celebrate wins but pivot earlier when needed. When it works, you know it.
Waiting for better computer use models, which are improving slowly but disappointingly. Surprising it's hard despite easy data availability. Language compresses easier into high-dimensional space than video feeds, but self-driving progress suggests mouse/clicking should be feasible. Coding agents became a workaround: script Excel, call APIs—coding improvements boosted Excel and commerce agents.
Legacy software needs computer use agents. For vibe coding platforms like Replit, testing apps (functional + UX/taste) requires augmenting models; Replit has a testing agent.
Continual learning is key (currently hacked via skill MD files). True on-the-job learning would enable deploying agents in orgs that improve specifically for them.
Company of the future: builders and salespeople. Sales evolves to evangelists/educators helping companies transform using the tech; humans are trusted for learning.
Builders operate at higher abstractions—computers abstracted human computation, software abstracted manual computer use, agents abstract software. Fully autonomous companies possible for some styles, but need business generalists: understand customers, economy, AI, vision.
Almost everyone is a founder. They wake up thinking how to make the company more successful, find problems, deputize agents to solve them.
Replit has a vibe coding resident team with vague mission: roam company, make it better. Examples:
- Support team (Zendesk): Built visualization/prioritization for queues by customer value/urgency; seat score improved.
- HR: Built internal platform for onboarding/benefits.
This generalist entrepreneur role is the future.
Keep Y Combinator in your library
Save the videos and channels worth coming back to, and find them again in one place.
