Replit's CEO On The Only Two Jobs Left In The Company Of The Future — Amjad Masad at YC Founder Firesides

Amjad Masad (Replit CEO) — The Only Two Jobs Left In The Company Of The Future

Replit is the leading no-code app builder for consumers and enterprise, letting anyone with an idea build real, deployed software using natural language. The company just raised a $400 million Series D at a $9 billion valuation.

In this episode of Founder Firesides, co-founder and CEO Amjad Masad sat down with YC’s Andrew Miklas to talk about Replit’s 10-year journey from browser IDE to vibe coding platform, why the people getting the most value aren’t traditional devs but founders and domain experts closest to the problem, and what Agent 4 unlocks with parallel agents, built-in design, and the ability to run your entire company on Replit.

0:28 – Anyone Can Build Software 2:14 – The Rise of AI-Native Builders 4:52 – Not Just Developers Anymore 7:18 – What People Are Actually Building 10:36 – How Replit Is Spreading Everywhere 14:02 – What You Can Build (and What You Can’t) 19:22 – YC, Growth, and Early Lessons 23:18 – From Vibe Coding to Autonomous Agents 29:44 – The Future: Everyone Becomes a Builder 36:12 – What Skills Matter Now

(Channel description, Y Combinator Founder Firesides upload, ~39 min runtime.)

A 39-minute YC Founder Firesides interview with Amjad Masad (co-founder & CEO of Replit) by Andrew Miklas (YC). Published 25 April 2026 — one day after Diana Hu’s Startup School session and two days after Garry Tan’s GStack demo, forming a three-day YC-batch-context cluster on the AI-native-company thesis from three complementary vantages (President / Partner / Founder of a portfolio platform vendor). ASR-cleaned for proper-noun consistency (Replit/Amjad Masad/vibe coding/Marc Andreessen/Paul Graham and Sam Altman/a16z/Zendesk).

TL;DR

The wiki’s first vendor-CEO source on vibe-coding-as-platform plus an explicit employment-future thesis from the same source. Four substantive contributions:

1. Replit as the vibe-coding-product first-mover (September 2024). Masad: “September 2024, Replit became the first like what’s called vibe-coding product where we abstracted away code entirely. So there’s like a coding agent behind the scene, but you’re just interfacing with AI using natural language.” This predates Andrej Karpathy’s December-2025 phase-change moment by ~3 months on the product side: Replit shipped the abstraction before the broader phase change Karpathy retroactively named. The wiki’s existing vibe-coding page tracks Karpathy’s coinage; this source adds the commercial-product-launch anchor that the floor-raising transition required infrastructure to materialise.

2. The AI-native-builder cohort is not who we expected. Two consecutive segments overturn the wiki’s prior assumption that vibe coding’s primary users are traditional developers experimenting on the side:

   • Traditional developers don’t get the most value from vibe coding — “developers like the pain… they like setting up things, they like configuring every aspect… It’s sort of like a craftsperson liking to build their own tools. People that are getting the most value tend to be the more tech-adjacent ones — product managers, designers, entrepreneurs.” This refines Nika’s PM-side worked example with a much broader audience claim and an explicit anti-developer-bias in the vibe-coding-platform design.
   • Domain experts can build what they need when they’re closest to the problem — the physical therapist (fascia-and-range-of-motion app after spending hundreds of thousands offshore), the pool-business SaaS founder (family-business domain), the sports-club founder (replacing MS-DOS-era software), the mom in Korea (rare-condition tracker for her child). Masad: “People who are closest to the problem can build the product they need.” The narrative is the floor-raising thesis substantiated with a non-Silicon-Valley audience — physical therapists and pool-business operators, not VC-backed founders.

3. Agent 4 = parallel agents + canvas + multi-modality + skills-call. The Agent 4 launch (April 2026, contemporary with the interview) packages four capabilities into a single release:

   • Parallel agents with merge-conflict resolution at the orchestrator level — the platform-side response to “the thing that sucks about autonomy is that you can put in a big prompt and you were kind of sitting back and just watching it work”. While one agent builds, the user designs / chats / kicks off other work asynchronously.
   • Canvas design surface — built-in design capability that lets the user explore the next page/feature while the build agent runs.
   • Multi-modal artefact projection — “when you make a mobile app, when you make a website, when you make a deck, when you make a video, all of that should have the context of your project.” One project → web app + mobile app (deployed to TestFlight) + deck + video — “so now you can run your entire company on Replit.”
   • Skills-call revolution — Replit’s MCP-and-integration layer: “as you’re talking to Replit and you’re like ‘I want to integrate Stripe’, we already built a set of skills and sometimes code… it’s sort of like Neo in the Matrix where you download a new skill and you’re like, ‘Oh, I know how to fly a helicopter now.‘” The skills-call framing parallels Tan’s GStack packaging at YC scale.

4. The company-of-the-future is made of builders and salespeople (37:00–end). The interview’s headline:

   • Salespeople are durable because “sales will be more like evangelists, more like educators — let’s help companies transform, let’s use the technology that we’re building to help companies transform. A lot of other companies will want to talk to someone, will want to learn from someone — they trust other humans.”
   • Builders are durable because “there’s always more to automate. There’s always… our job will continuously get higher and higher level.” Masad reaches back to the human-computers→machine-computers→software-engineers→agentic-engineers abstraction-ladder argument as evidence.
   • The internal vibe-coding-resident team as the worked example. Replit has a small team with a deliberately vague mission — “go around the company and make it better.” They went to the support team, found a CSAT-pulling-down ticket-prioritisation gap, built a priority-queue tool; CSAT went up. Then to HR, found onboarding pain, built an internal HR platform. “And I think that kind of role is more where the future is headed and more and more people inside the company will be more generalist entrepreneurs that are trying to make the business successful.” The structural claim: everyone-is-a-founder is the natural endpoint of vibe-coding-as-organisational-baseline.

Plus a fifth substantive thread on what skills will matter (29:44–end): playing-with-tools / having-the-playful-mindset, being plugged in / online (“at some point it was kind of a drawback to be someone constantly reading the news, but actually now it’s very important to know what’s happening, what’s coming down the line”), not giving up (“the thing that you try to do today that AI cannot do, try it again in a month” — the unbounded-by-current-capability mindset), idea generation as residual durable skill, taste (Masad’s specific call-out: “things that are functionally right but UX is not very well” — taste is currently the bottleneck for computer-use agents).

What was actually ingested

Full ~39-minute episode transcript (auto-generated ASR, ~382 raw segments). Light cleanup for proper-noun mistranscriptions (Replet → Replit, Msad → Masad, Vive coding / Wipe code → vibe coding, Mark and Dreon → Marc Andreessen, openclaw → OpenCode). No chapters in the YouTube data — the description-level timestamps are the navigation surface. Transcript ends cleanly with both speakers’ sign-off.

Operational mechanics — the four-layer abstraction Replit climbed

Masad’s 10-year history of Replit is a four-layer abstraction-removal ladder; the interview makes this structure explicit:

Year	What Replit removed	Layer that was hard
~2016–2018	The development-environment setup	Setting up a stack on your laptop
~2019–2021	The deployment setup	Hosting your app, scaling, ops
2024-09	Writing code itself	Translating intent → syntax
2026-04 (Agent 4)	The boundary between artefacts	One project = web + mobile + deck + video

Each step is a rapid-prototyping move (digital-seizing/rapid-prototyping) but also a business-model refresh (strategic-renewal/business-model): the value proposition mutates from “better IDE” to “a place where ideas become deployed software” to “run your entire company on Replit.” The CEO is describing the firm’s value-creation logic shifting underneath itself.

The six-month-step-change roadmap discipline

Masad: “We thought that broadly the AI capabilities have massive step changes twice a year. So if you think about 2025, there was the vibe-coding revolution in early 2025 and then the autonomous revolution in late 2025 with Opus 4.6… mid-2024 Claude / Sonnet came out and for the first time we could generate a lot of code… late 2024 we started seeing initial signs of the labs doing long-horizon reasoning. And so from that observation we’re like, OK, we want to align our roadmap to AI capabilities. And so every six months we release a new agent version.”

The product-strategy claim: a coupled-to-foundation-model-step-changes release cadence is the right discipline for vibe-coding-platform vendors. Two consequences:

1. Predictive over-build: Agent 3 was “the most autonomous agent on the market” in September 2025 when “true autonomy didn’t arrive until like November / December” — the platform had to ship the long-running-container infrastructure ~3 months before the model side was ready. This is contextual/external-triggers-driven roadmapping in its strongest form: the platform commits capital to capability bets before the capability lands.
2. Risk of mis-timed bets: Computer-use agents are a “mystery” — Masad expected them to mature faster (“if we can make progress on self-driving, surely we can make progress on moving the mouse”). The wiki should track whether the bi-annual cadence holds or breaks.

Convergence with the wiki’s existing corpus

Source	What it adds relative to Masad
Karpathy at Sequoia AI Ascent	The coined term and the December-2025 phase-change. Masad provides the product-launch anchor (September 2024) that predates Karpathy’s framing by ~3 months on the platform side. Read together: Karpathy named what Masad shipped
Nika 2025	The PM-side worked example. Masad generalises the audience claim from PMs to all tech-adjacent non-developers + domain experts. Together they establish a floor-raising audience pipeline from individual PM workflow → cross-functional non-developer audience
GStack	YC-president-level worked example of vibe-coding-with-discipline. Masad provides the platform-vendor parallel: skills-call revolution = GStack-style packaging at the Replit-platform layer
Startup School	YC-partner-level prescription (“everyone builds and ops support sales”; closed-loop companies; software factories). Masad gives the founder/CEO-vantage worked example of Hu’s structure: Replit’s internal vibe-coding-resident team is literally the closed-loop pattern in operation
Jassy May 2025	Pre-coinage CEO anchor for the floor-raising thesis. Masad’s “so much software to build in the world… when anyone can make software, a lot of parts of the economy is just going to improve” operationalises Jassy’s prediction with a vendor doing the building
The Daily	75-developer survey anchor at small-startup (~100% AI-written) and Google (~40–50%) scales. Masad gives the platform-side cross-audience anchor: Replit’s users are not the developers Thompson surveyed — they’re physical therapists, pool-business owners, sports-club ops
vibe-coding concept page	This source adds: (a) the platform-launch date anchor; (b) the anti-developer-bias design observation; (c) the post-prompting product direction; (d) the vibe-coding-resident team as a worked organisational pattern

Linked entities and concepts

• Amjad Masad — co-founder & CEO of Replit; the substantive speaker for ~95% of runtime. First wiki appearance as a source-anchored figure; per the second-source rule, entity page deferred (single source). If a second source cites Masad — likely given the platform-vendor cadence — promote on that ingest.
• Replit — vibe-coding-platform vendor; $400m Series D at $9bn valuation announced ahead of this interview. First wiki appearance as a primary source subject (previously referenced only in passing in vibe-coding). Single source; entity page deferred.
• Andrew Miklas — YC interviewer. Single-source mention; deferred.
• Y Combinator — channel/author. Third source from the YC channel (Tan + Hu precede it). Entity page already exists with source_count: 2 — this ingest bumps it to 3 and bumps last_confirmed to 2026-05-14.
• Marc Andreessen / a16z — passing mention (a16z led Replit’s seed round). Single-source mention; deferred.
• Paul Graham / Sam Altman — passing mention (saw Replit on Hacker News, invited to YC). Deferred.
• Peter Levels (or “Levelvel’s type of entrepreneur” — likely refers to Pieter Levels, the indie-maker reference) — single-source mention; deferred.
• Agent 4 — Replit’s April 2026 agent release. First wiki mention; not a stand-alone wiki concept.

Dangling (single-source mentions, deferred): Amjad Masad, Replit, Andrew Miklas, Marc Andreessen, a16z, Pieter Levels.

Body-wikilink coverage for all relationships: targets is satisfied through the TL;DR and the Convergence table.

Open questions raised

• Does the bi-annual capability step-change cadence hold? Masad’s mid-2024/late-2024/early-2025/late-2025 narrative is post-hoc clean; the wiki should track whether 2026 H1/H2 fits the pattern (Agent 4 in April 2026 implies an Agent 5 in October 2026).
• The “computer-use agent mystery”. Masad expected computer use to mature faster than language; it hasn’t. This is a candidate for the wiki’s open-questions cluster — paired with the HF agentic-evals workshop’s sim-to-real-gap discussion, it suggests the bottleneck is environment representation, not raw capability.
• Will the vibe-coding-resident pattern generalise beyond Replit? Masad describes it as a normal internal team structure. The wiki should track whether Hu’s YC-batch is seeing this pattern propagate — second sources documenting the generalist-entrepreneur-inside-a-bigger-company role at non-Replit firms would close the loop.
• Post-prompting at scale. Masad: “I think we’re headed to a post-prompting world… at some point perhaps Agent 5 or maybe sooner, you should be able to tell Replit ‘every day build me a SaaS company and try to market it and see what works.‘” If post-prompting becomes the platform mode, the eval discipline (cf. Husain and HF Agentic Evals) becomes more critical, not less — the wiki should track whether the eval-as-offense framing scales to the post-prompting layer.
• Two-jobs claim falsifiability. “The company of the future is made of builders and salespeople” is a strong, falsifiable claim. The wiki should keep an eye on whether other durable role categories surface across the corpus (e.g. trust/relationship roles, compliance/risk, taste/curation) and whether Masad’s two-category framing holds up.

Related sources

• 2026-04-29-andrej-karpathy-from-vibe-coding-to-agentic-engineering — the term itself; the December-2025 phase change. Karpathy named what Masad shipped.
• 2025-12-01-marily-nika-pms-who-use-ai-will-replace-those-who-dont — PM-side worked example of floor-raising; Masad generalises the audience.
• 2026-04-23-tan-yc-how-to-make-claude-code-your-ai-engineering-team-gstack — YC-president platform packaging; parallel to Masad’s skills-call revolution at the Replit-platform layer.
• 2026-04-24-hu-yc-how-to-build-a-company-with-ai-from-the-ground-up — YC-partner prescription that Replit’s internal vibe-coding-resident team operationalises.
• vibe-coding — concept page this source materially extends with platform-launch date, audience refinement, and post-prompting direction.
• ai-employment-effects — concept page that the builders-and-salespeople thesis directly addresses.
• durable-skills — concept page that Masad’s skills-that-matter list operationalises at individual-posture level.