Tun Shwe is Head of AI at Lenses.io, building tools that connect Apache Kafka and real-time data infrastructure with production AI agents. His work tackles a central problem for autonomous systems: providing current operational context without exposing sensitive information or granting excessive access.
Earlier, Shwe worked as a data engineer and vice president of data at high-growth startups, leading teams developing analytics platforms and data-intensive AI applications. He co-organized PyData London, co-founded PyData Cornwall and contributed to Quix Streams, an open-source Python stream-processing library.
At Lenses, he helped develop Lenses MCP for Kafka, an open-source Model Context Protocol server connecting agents with streaming environments. His premise: stronger models cannot overcome stale operational context when deployments, schemas, incidents and pipelines change continuously. Agents need live information bounded by permissions, governance and auditability.
- Secure-by-design agent interfaces. In an AI Engineer Europe session with Jeremy Frenay, Shwe argues that interface design determines security: consolidate tools around specific outcomes, validate inputs with schemas or Pydantic, write unambiguous descriptions and return only essential information. These measures limit tool poisoning, command injection and sensitive-data exposure.
- The MCP security cliff. Moving a server from one developer’s laptop to shared infrastructure introduces authentication, transport security, scoped authorization, token management, rate limiting and audit requirements simultaneously. Shwe examines these demands in his writing on production security for MCP servers.
- The agentic software factory. His model for agentic software delivery extends automation beyond writing code into infrastructure, data pipelines, incident response and governance, with durable execution, human approval and measurable cost per completed task.
