Anush DSouza is a senior product manager at Heroku focused on the infrastructure and developer tools behind production AI applications. His work on Heroku Managed Inference and Agents, the Heroku MCP Server, and isolated code execution helps developers connect language models to software and organizational data without surrendering control over deployment, permissions, or compute.
Earlier in his career, DSouza worked across software engineering and product management at Microsoft, Goldman Sachs, and Sprinklr. He holds a master’s degree in computer science from the University of Southern California.
In 2025, DSouza coauthored the introduction of the official Heroku MCP Server, which lets AI clients perform platform operations through Heroku’s existing command-line tooling. The subsequent Heroku Remote MCP Server extended those capabilities beyond a developer’s computer through OAuth-authenticated agent connections.
At AI Engineer World’s Fair 2025, DSouza joined Julián Duque for a workshop on managed inference and agents. DSouza emphasized operational challenges that follow an initial prototype: model selection, scaling, trusted tool execution, and integrating inference, compute, databases, and Model Context Protocol extensions into one developer workflow.
- Secure agent interoperability. DSouza’s MCP work makes infrastructure accessible to tools including Claude, Cursor, and Agentforce while retaining authentication and explicit access boundaries.
- Isolated code execution for agents. His agent sandbox architecture uses temporary one-off dyno sandboxes to run generated code. Processing larger datasets outside the model’s context window allows agents to return compact results without repeatedly passing bulky intermediate data through inference.
- Production-quality retrieval. His writing on AI search architecture and data-aware applications with LlamaIndex addresses a practical weakness of vector similarity alone: semantically related documents do not necessarily answer the user’s question. He emphasizes retrieval design and access to relevant private data.
More recently, DSouza has helped expand Heroku’s managed model catalog and deployment options, continuing to simplify how developers configure and operate AI applications.
