Claude

How Emergent is making app building more accessible with Claude

Emergent reached $100M ARR in eight months, with 70–80% of users having never written a line of code. In this conversation, Co-founder and CEO Mukund Jha sits down with Anthropic's Carly Ryan to talk about the technical decisions behind building on Claude and how Emergent thinks about durability in a category moving this fast. Read the full story: claude.com/customers/emergent

MJMukund JhaguestCRCarly Ryanhost

May 13, 2026·16m·Watch on YouTube ↗

📖 CHAPTERS

1. 1

   Emergent’s mission: AI-built apps for the “million niches” of small business

   Mukund frames Emergent around a macro thesis: small businesses drive a huge share of global jobs and GDP but have historically lacked software tailored to their niche needs. AI changes the economics by enabling bespoke software at near-zero marginal cost, making custom tools accessible beyond traditional SaaS constraints.

2. 2

   From twin founders to YC: the initial bet on automating software engineering

   Mukund recounts the founding story with his twin brother and their long-standing obsession with automation in software engineering. The first YC direction focused on automating software testing, motivated by his prior experience running a large engineering organization where testing was a shipping bottleneck.

3. 3

   The pivotal insight: verification loops enable long-running coding agents

   While building automated testing, Emergent realized they needed deeper agent and infrastructure capabilities. Mukund describes a key realization: if you solve verification/feedback loops, agents can run longer and tackle more complex software tasks reliably.

4. 4

   Pivot from enterprise to democratized app-building for non-technical users

   Emergent initially pursued enterprise customers but found adoption slow. Internally, non-technical teammates were getting outsized value from the agents, revealing a stronger opportunity: democratize software creation so domain experts can build without traditional engineering skills.

5. 5

   “Production-ready, not demos”: building durability through testing, security, deployment and reviews

   Mukund contrasts Emergent’s approach with tools optimized for quick prototypes or front-end demos. Emergent prioritized the unglamorous but critical parts of real software delivery—testing, linting, security, deployment, and code review—so users can ship business-critical apps.

6. 6

   Why Claude (Sonnet → Opus): quality, instruction-following, and long-horizon agents

   Mukund explains why Claude became the default model choice: strong instruction-following and coding performance enabled the team to envision longer-running agents earlier. Emergent optimizes for correctness and outcomes over latency or raw cost, with Opus as a “workhorse” paired with tight feedback loops.

7. 7

   Owning the whole stack: containers, snapshots, harness quality, and multi-agent orchestration

   Emergent’s core technical learning was that automated software engineering requires controlling the full stack end-to-end. Mukund details building custom container tech atop Kubernetes (including snapshotting) and the importance of the “harness” around the model—memory, context management, caching, and agent-to-agent coordination.

8. 8

   Learning from production: logs, long-term memory, and self-improving agents

   Because Emergent hosts and deploys apps, it can capture real production failures that don’t show up in development. Those signals feed back into development, powering long-term memory and faster future fixes with fewer tokens and higher accuracy across apps and library upgrades.

9. 9

   Model selection when users can’t debug: optimize for reliability and success metrics

   Emergent carries the burden of correctness because many users can’t debug generated code. Mukund describes measuring success via deployability and improving deployment rates dramatically by tightening loops and increasing agent reliability, while choosing frontier models for reasoning over cost or latency constraints.

10. 10

    What’s hard to replicate: vertical tech stack focus + infrastructure + real business requirements

    Mukund argues the market is expanding rapidly, but Emergent’s defensibility comes from customer closeness and the difficult “last-mile” engineering beyond code generation. He highlights business-critical needs like payments, authentication, large files, scalability, and end-to-end parity between dev and production.

11. 11

    Quality layers around the agent: refactoring, pre-/post-deployment security checks

    Emergent adds specialized agents to improve maintainability and safety after initial generation. Mukund describes refactoring and security-focused agents that scan for vulnerabilities and key leakage, reinforcing the “durable product” promise for business owners.

12. 12

    Carly’s moat framework and Emergent’s pace: rebuild for the next model frontier

    Carly outlines defensibility beyond model intelligence—user understanding, proprietary assets, compliance/regulation, and human trust. Mukund adds that Emergent has rewritten systems multiple times to reimagine what’s possible with each new model generation, especially as long-horizon coordination improves.

13. 13

    Who uses Emergent and what they build: domain experts shipping real apps (EquiMind story)

    Emergent’s biggest surprise was the user base: business operators and domain experts—not just semi-technical PMs or designers. Mukund shares a concrete example of a clinical psychologist/equestrian coach who shipped an app after years of waiting and failed dev-shop attempts, illustrating the platform’s leverage.

14. 14

    The roadmap: from building software to “autonomous businesses” with Wingman

    Looking ahead, Mukund anticipates a shift toward automating broader business operations as long-horizon agents mature. Emergent plans to expand beyond app-building into automating finances, operations, sales, and marketing via a new product called Wingman.