scripod.com

⚡️ Ship AI recap: Agents, Workflows, and Python — w/ Vercel CTO Malte Ubl

In this episode, Malte Ubl, CTO of Vercel, shares how the company is redefining AI-powered development with a focus on durable workflows, production-ready agents, and developer experience. The conversation reveals how Vercel’s engineering philosophy shapes its approach to building scalable, secure, and practical AI systems that integrate seamlessly into real-world software workflows.
Vercel is advancing AI engineering through its Workflow Development Kit, enabling serverless functions to pause, resume, and retry—ideal for complex agent orchestrations and human-in-the-loop processes. Their AI SDK 6.0 uses a low-level, flexible architecture that avoids over-abstraction, allowing developers to adapt quickly to new AI patterns. By 'dogfooding' their tools, Vercel built internal agents for DevOps, code review, and lead qualification that handle repetitive tasks with high reliability. These agents reduce alert fatigue by investigating anomalies using real production data before escalating. The 'Agent on Every Desk' program helps companies deploy their first agents with templates and support. Vercel also launched native Python support for Flask and FastAPI, emphasizing zero-config deployment. Security is prioritized through infrastructure that enforces safety regardless of AI-generated code quality, separating access controls from application logic. The vision centers on empowering individual contributors and helping organizations evolve in the AI era without sacrificing trust or maintainability.
02:46
02:46
Workflows allow serverless functions to wait, resume, and retry automatically without extra cost.
14:21
14:21
Letting the model take control enables emergent behavior in coding agents
19:14
19:14
Agents can investigate production anomalies without waking engineers, enabling aggressive tuning of detection systems.
29:33
29:33
Vercel helps companies build their first agent, guides on the second, and expects the third to be built independently.
39:42
39:42
Vercel builds security models assuming developer and AI incompetence to enforce safe defaults.