Replit's CEO On The Only Two Jobs Left In The Company Of The Future

Guest & Company Introduction [0:04-0:25]

The guest is Amjad Msad, CEO and co-founder of Replit, described as the leading no-code app builder for consumers and enterprise. Replit recently raised a Series D round of $400 million at a $9 billion valuation.

What Replit Is & Its Mission [0:26-2:07]

Replit's ambition is that anyone who can read and write can come in with an idea and leave with a deployed, hosted app that scales. Initially, Replit solved development and deployment environments. In September 2024, it became the first 'vibe coding' product, abstracting away code entirely with a coding agent. Agent 4 added design interactions via a canvas. The goal is to create a natural place where ideas turn into real, secure, scalable software.

Target Users & The Shift Away from Developers [2:08-5:43]

Amjad started coding young and felt developer tools added accidental complexity. Replit initially aimed to 'make programming accessible' and later 'create a billion new developers.' A key insight was that traditional developers 'like the pain' of setup, so Replit shifted to target 'tech-adjacent' users: product managers, designers, and entrepreneurs. In 2023, Replit explicitly decided not to go after traditional developers but to build for 'creators'—AI native developers.

What People Are Building on Replit [6:28-11:02]

Three main categories: personal software, enterprise software, and entrepreneurial products. Examples include a physical therapist building a health tech app (replacing costly offshore development), a pool maintenance SaaS, sports club software replacing MS-DOS, a mom managing her kid's rare condition, and a 'chore hero' iPad app. Enterprise use cases include Whoop using Replit to try more ideas (order of magnitude increase) and RevOps building internal tools like quote configurators, saving millions on SaaS.

How Replit Finds Users & Sales Motion [11:05-13:30]

The shift of empowered decision-making now includes product managers and operations managers. Consumer use overlaps with work use. Go-to-market is product-led growth with sales assistance. A champion brings Replit to work, and Replit helps convince leadership via hackathons. Top-down enterprise sales also occur, where Replit wins on trust, security, and compliance.

Limits of Replit & What It Can Build [13:33-16:27]

Amjad confidently states entrepreneurs can build SaaS, consumer, or automation products on Replit, but not new cloud platforms or ML systems. 'Replit native agencies' emerge, offering 60-70% cheaper services. Internal tools and automations are common. Replit integrates with partners and uses MCPs and 'skills' to expand capabilities.

Building a Community for Non-Developers [16:29-19:25]

Community building requires showing what's possible. DevRel acts as educators, documentation is simpler, and video content is key. The agent serves as a brainstorming partner. Replit moved from pure prompting to a canvas with visual shortcuts. Enterprise engagement includes hackathons. The ideal champion is entrepreneurial, resourceful, influential, and has a founder mindset.

YC Experience & Influence [19:28-23:01]

Key realization: how much can be done in 3 months. Sam Altman advised hyperfocus. Replit entered YC as a simple command line and exited with web dev, hosting, and IDE features. Now, agent releases involve four-week sprints with 24/7 work. Another lesson: compound growth (7% week-over-week). YC opened doors for fundraising; a16z led the seed round after a breakfast with Marc Andreessen. Amjad credits YC for their success.

Agent 4 & The Future of the Product [23:11-23:30]

Agent 4 is the fourth major revision. Starting in 2024, Replit predicted a massive step function in AI capabilities.

Timeline of AI and Agent Capabilities (2024-2025) [23:39-24:14]

Early 2025 saw the 'vibe coding revolution.' Late 2025 had an 'autonomous revolution' with Opus 4.6. Mid-2024, Cloud Sonnet was the first model to generate a lot of code. Late 2024 showed initial signs of long horizon reasoning.

Product Strategy and Release Cycle [24:20-24:36]

Replit aligns its roadmap to AI capabilities and releases a new agent version every six months, both predicting and pushing the edge of what's possible.

Agent 3 (Autonomy Focus) [24:38-25:21]

Agent 3 was the most autonomous agent on the market, running for hours to build software from a big prompt. Replit rewrote the back end for long-running containers. True autonomy arrived later, but the agent demonstrated the direction.

Agent 4 (Parallelism and Asynchronicity) [25:25-27:16]

Agent 4 introduced parallel agents to avoid users watching the agent work. Users can design, kick off other work, and chat while the agent builds. Replit built a canvas for design, solved merge conflicts for multi-agent architecture, and enabled teamwork by forking VMs. The orchestrator subdivides tasks, and the canvas shows other cursors for live collaboration.

Agent 4 (Multi-Product Context) [27:22-28:04]

Agent 4 allows building mobile apps, websites, decks, or videos with project context. Users can say 'make a mobile app' from a web app, and the system deploys to both web and TestFlight/Android. The claim: you can run your entire company on Replit.

Skills for the Future (Post-Prompting World) [28:23-30:52]

We are headed to a post-prompting world where users give high-level goals like 'optimize my marketing funnel.' By agent five, users should be able to tell Replit to build and market a SaaS company daily. Key skills: understanding what's possible, having a playful mindset, staying plugged in, not giving up (try again in a month), and idea generation. Being creative and figuring out what the world needs is crucial.

Lessons Learned / Advice for Starting Over [31:00-31:57]

Mistakes include not prioritizing culture early (leading to a reset and layoffs) and deluding oneself about product-market fit. Getting any user or money is an achievement, but true product-market fit is explosive. The speaker notes periods where they should have changed directions earlier.

AI Technology: Computer Use Models & Limitations [32:03-34:42]

Computer use models are slowly improving but disappointing; language is easier to compress than video. Coding agents are a workaround for computer use. Replit spends time augmenting computer use models for testing. Continual learning is not yet unlocked; current hacks involve writing to files. True learning on the job remains far away.

The Company of the Future [35:14-38:57]

The future company consists of builders and sales people. Sales will become more about evangelism and education. Builders will continuously automate higher-level tasks. The speaker envisions a company where almost everyone is a founder, deputizing agents to solve problems. Replit's 'vibe coding and resident team' exemplifies this, building internal tools like support queue prioritization and HR platforms, improving metrics like seat score.