Abhi Aiyer on building mastra and the future of AI agent frameworks
Michael Grinich interviews Abhi Aiyer from Mastra about building open-source AI agent frameworks, developer tooling, and the evolving agentic ecosystem.
Michael Grinich sits down with Abhi Aiyer from Mastra at HumanX 2026 to discuss the challenges of building an open-source AI agent framework, why developer experience matters for agentic workflows, and where the ecosystem is headed.
At HumanX 2026 in San Francisco, WorkOS CEO Michael Grinich caught up with Abhi Aiyer from Mastra — the open-source TypeScript framework for building AI agents and workflows. They talked about what it takes to build developer tooling for an era of autonomous AI systems, how Mastra approaches the agent framework problem, and why the current moment feels a lot like the early days of React.
What is mastra?
Mastra is an open-source TypeScript framework designed to help developers build AI agents, workflows, and RAG pipelines. Rather than forcing developers to learn a new paradigm from scratch, Mastra meets them where they already are — in the TypeScript toolchain — and gives them composable primitives for building applications that run autonomous AI workflows.
The framework handles the hard infrastructure pieces: tool calling, memory, structured workflows, and integrations with model providers. Developers get to focus on the logic of their agents rather than reinventing plumbing.
Why TypeScript for AI agents?
Most AI agent frameworks started in Python. Mastra made a deliberate bet on TypeScript. Aiyer explains that the reasoning is straightforward: a large share of production web applications and APIs are already built in TypeScript or JavaScript. Forcing those teams to context-switch into Python just to add autonomous agent capabilities creates unnecessary friction.
By staying in TypeScript, Mastra lets full-stack developers build agents with the same language, toolchain, and deployment patterns they already know. That's not a minor ergonomic win — it removes the need to maintain a separate Python service, which simplifies deployment and reduces operational overhead.
The agent framework space
Aiyer draws a parallel to the early JavaScript framework wars: there's a proliferation of approaches, strong opinions about architecture, and a lot of experimentation happening simultaneously.
What separates frameworks that last from those that don't? Aiyer points to a few things:
Building in the open
Mastra's open-source approach shapes how the team builds. Aiyer described the feedback loop with the developer community as one of Mastra's biggest advantages — issues, PRs, and feature requests from real production use cases drive the roadmap more than internal speculation about what developers might want.
This tight feedback loop is common across successful developer tools. The frameworks that win aren't necessarily the ones with the most features on day one. They're the ones that listen, iterate fast, and earn trust through transparency.
What's next for AI agent tooling
Aiyer is clear-eyed about where things stand: we're still early. The patterns for building reliable, production-grade agents are still being established. Memory architectures, evaluation frameworks, and multi-agent coordination are all active areas of research and development.
Agents are moving from demos to production workloads, and the tooling needs to keep pace. Mastra is betting that the developers who build the next generation of AI-native applications will want a framework that feels familiar, stays out of their way, and gives them full control when they need it.
This interview was recorded at HumanX 2026 in San Francisco.