Temporal's Strategy: Leveraging Durable Execution for AI Agents and Internal Workflows
This article explores how Temporal, a durable execution platform, is internally adopting AI to enhance workflows and enable AI agents, while also positioning its core technology to solve reliability challenges for long-running AI agentic systems. It highlights the company's cultural shift towards AI adoption, the architectural benefits of using durable execution for critical internal processes, and the challenges of managing AI costs and 'AI slop'.
Read original on The New StackDurable Execution as the Foundation for AI Agents
Temporal's core offering is durable execution, a mechanism designed to keep long-running software processes alive through failures, crashes, and outages. The article emphasizes that this same problem space—ensuring reliability and fault tolerance for asynchronous, long-lived, and mission-critical operations—is precisely what teams building production AI agents now face. Temporal positions its platform as a solution to make agentic AI reliable by providing a robust underlying infrastructure that can manage state, retries, and error handling for complex, multi-step AI workflows.
Internal AI Adoption and Workflow Transformation
Temporal is actively implementing an "AI for Temporal" strategy, mandating company-wide AI adoption not just for engineering but for all departments. This involves rethinking existing workflows and leveraging AI tools to improve efficiency. For instance, their security team transformed its vulnerability management from a ticket-based system into a durable workflow built on Temporal itself, named "Deputy." This workflow automates the entire lifecycle: finding flaws, generating fixes, opening PRs, merging, running CI, and rolling out updates. This demonstrates how durable execution can be applied to create highly resilient and automated internal systems.
Automated Security Workflow with Temporal
The "Deputy" internal tool automates the entire security vulnerability lifecycle, from detection to deployment of fixes, using Temporal's durable execution. This significantly reduces friction and accelerates resolution times, shifting the security team's focus from issue filing to fix shipping.
Key AI Adoption and Management Practices
- Mandated but Flexible Tooling: While AI adoption is mandatory, specific tools are not. Engineers gravitate towards tools like Claude Code and foundation models from OpenAI/Anthropic. Open-weight models are currently not broadly approved due to security and privacy concerns.
- Cost Management: AI costs (primarily from coding agents) increased fivefold. Temporal uses tools like Vantage to track vendor costs, sets hard spending limits, and reviews outliers bi-weekly. However, it avoids employee usage leaderboards to prevent counterproductive incentives.
- Addressing 'AI Slop': The increased ease of generating content with AI has led to a rise in "sloppy work." Temporal addresses this culturally by making it acceptable to return AI-generated content if it hasn't been self-reviewed, emphasizing quality over raw output.
The article also highlights the development of an internal platform built on Temporal where non-technical employees can describe an app to Claude Desktop and deploy it with a single click. This platform incorporates guardrails, identity, security, and uses Temporal's own model routers to prevent vendor lock-in, showcasing an architectural approach to empower low-code users with robust, AI-powered application building.