

AI Infrastructure Night
AI Infrastructure Night brings together founders and engineers building the systems behind production AI.
Four technical talks on building, deploying, and operating production AI systems.
👽 Join the Alien Slack Community https://alien.dev/slack
Connect with engineers, founders, and technical leaders building AI products and infrastructure.
✨ Speaker lineup
5:00–6:00 PM · Arrival & Networking Food, drinks, and introductions.
6:00–6:15 PM · Constal, Sidhartha Mani, Founder & CEO
Beyond One Sandbox Provider: Routing, Failover, and the Tradeoffs
Getting your first sandbox is easy. Designing an agent system that can use more than one provider is where the interesting engineering begins.
This talk explores how to choose where a sandbox runs, respond when a provider runs out of capacity, and balance cost, performance and availability. We’ll distinguish retrying a provisioning request from recovering work already in progress, and discuss where adding another provider helps versus where it adds complexity.
A practical look at multi-provider sandbox infrastructure for developers building agents that run code.
6:15–6:30 PM · Chalk AI, Elliot Marx, Co-founder
The New AI Infrastructure Stack: Built for Agents
AI agents are changing how software is built. As agents take on more complex tasks and operate in parallel, they need a new infrastructure layer that provides isolated compute, secure access to tools and data, and environments that can be provisioned and managed programmatically. In this talk, we’ll explore why traditional development environments are becoming a bottleneck and what it takes to build infrastructure designed for autonomous agents. We’ll examine how scalable, secure sandboxes and programmable development environments are emerging as foundational building blocks of the next generation of AI infrastructure.
6:30–6:45 PM · Alien, Aditya Mahamuni, Ambassador
How do you operate software inside infrastructure you don't control?
Deploying into a customer’s cloud once is straightforward; upgrading and debugging it across hundreds of opinionated environments is not. This talk covers drift, permissions, partial failures, upgrades, rollbacks, observability, and supporting Terraform, Helm, and internal platform tooling without direct access.
6:45 PM-7:00 PM · Datastrato, Mark Hoerth, Product Lead and Solutions Architect
Governing AI Agents’ Access to Data: The Catalog as Control Plane
AI agents now query enterprise data independently, in parallel, and across lakehouses, warehouses, and streaming systems. Each system enforces access its own way, and none of them knows which agent is acting for which person. As a result, teams either give agents over-privileged service accounts or the agents can’t reach the data they need. Using Apache Gravitino as the example, the talk covers how a federated metadata catalog becomes the control point for agent data access. It covers identity that carries the human principal through to the agent, privileges and row and column restrictions defined once and enforced across engines, short-lived credentials vended per request, and an MCP gateway with semantic models that give agents business context instead of raw schemas.
7:00–7:30 PM · Networking Continue the conversations with speakers and fellow attendees.