Cornelia Davis is a principal technologist at Temporal, former chief technology officer of Weaveworks, and author of Cloud Native Patterns. She applies decades of experience in distributed systems to a central problem in AI engineering: keeping autonomous applications running when workers crash, networks fail, external services stall, or human approval takes days.
Davis earned computer science degrees at California State University, Northridge, and studied at Indiana University. She developed imaging systems at Hughes Aircraft, worked on web-based collaboration at eRoom Technology, and moved through Documentum into EMC’s corporate technology organization. After building Cloud Foundry service brokers and an early BOSH release, she joined the Cloud Foundry team in 2013 and later held platform-engineering and transformation leadership roles at Pivotal.
Her 2019 book, Cloud Native Patterns, distilled that experience into principles for change-tolerant software, including service discovery, retries, resilient routing, and independent deployments. As Weaveworks chief technology officer, she advanced GitOps and cloud-native operations, emphasizing repeatable infrastructure management and automated reconciliation across the Kubernetes ecosystem.
At Amazon Alexa, Davis led Structured Knowledge Skills, giving enterprise customers natural-language access to structured data through knowledge graphs and question-answering. She subsequently worked as a technology fellow and vice president of product at Spectro Cloud, focusing on Kubernetes management, edge deployment, and intermittent connectivity. At Temporal, she brings those platform, conversational-AI, and infrastructure disciplines together in work on production-ready agent systems.
- Durable execution for AI agents: Davis uses workflows, activities, queues, and event-sourced execution to preserve completed model calls, tool results, and application state across failures. Her OpenAI Agents SDK demonstration shows a crashed worker resuming a multistep weather-query workflow without repeating successful inference. The integration exposes Temporal activities as agent tools through
activity_as_tool.
- Long-running human-in-the-loop workflows: Davis distinguishes a business process’s logical lifetime from any individual server process. Approvals can wait hours or days without keeping workers alive, then resume from recorded state. She also stresses idempotent activities because a retried external operation might already have succeeded before its response was lost.
- Durable MCP Tasks: Her asynchronous Model Context Protocol implementation tracks purchase orders, ERP invoice processing, human approvals, and concurrent tasks across disconnected clients or servers. She favors explicit task updates and persisted client-side identifiers over stateful task discovery and long-lived response channels. At larger scale, she argues that targeted notifications can avoid polling every outstanding task.
- Composable micro-agents: Davis applies microservices thinking to narrowly scoped agents orchestrated through code, parallel execution, or framework handoffs. She distinguishes separate agentic loops from handoffs that change the context within an existing loop—a practical difference when designing recovery, ownership, and state management.
