Descriptions:
This Anthropic-produced interview features Makund, CEO and co-founder of Emergent, in conversation with Karly from Anthropic’s applied AI team. Emergent — co-founded with Makund’s twin brother Madhav — started in Y Combinator focused on automated software testing before pivoting to general-purpose coding agents, eventually launching a product aimed at enabling non-technical users to ship production-ready software without engineering teams.
Makund details the technical foundations Emergent built along the way: proprietary container infrastructure, a verification loop that allows agents to run longer, multi-agent communication systems, and what the team calls long-term memory — an architecture that allows agents to learn not just within a session but across all applications built on the platform. He credits this accumulated feedback loop, combined with owning the full stack through to hosting, as the reason Emergent’s production deployment rate climbed from 84% to 98%.
On model selection, Makund explains that Emergent benchmarks itself against dev shops charging $250,000+ for comparable projects, making users relatively insensitive to latency or cost — which is why the company consistently prioritizes the highest-reasoning frontier models over cheaper or faster alternatives. He specifically credits Claude Sonnet’s early instruction-following and coding capabilities as foundational to Emergent’s long-running agent architecture. The video offers a candid look at how a fast-growing AI app builder thinks about model choice, memory systems, and serving the small business market at scale.
📺 Source: Claude · Published May 13, 2026
🏷️ Format: Interview
