⚡️ Agents, Workflows, and Python: Ship AI recap w/ Vercel CTO Malte Ubl
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this conversation with Malte Ubl, CTO of Vercel, we explore how the company is pioneering the infrastructure for AI-powered development through their comprehensive suite of tools. (01:00) Malte shares insights into Vercel's philosophy of "dogfooding" - never shipping abstractions they haven't battle-tested themselves - which led to extracting their AI SDK from v0 and building production agents that handle everything from anomaly detection to lead qualification. (11:57)
The discussion dives deep into Vercel's new Workflow Development Kit, which brings durable execution patterns to serverless functions, allowing developers to write code that can pause, resume, and wait indefinitely without cost. (05:20) Malte explains how this enables complex agent orchestration with human-in-the-loop approvals through simple webhook patterns, making it dramatically easier to build reliable AI applications.
Key themes covered include:
- Vercel's strategic approach to AI infrastructure, including their DevOps agent that automatically investigates production anomalies and their "agent on every desk" program for forward deployment
Speakers
Malte Ubl
Malte Ubl is the CTO of Vercel, where he leads the company's technical vision and infrastructure development. Prior to Vercel, he worked at Google on Search, AMP, and internal tools like Wiz, contributing significantly to web technologies that power much of the modern internet. He joined Vercel about 3-4 years ago, before the ChatGPT era, and has been instrumental in transforming the company to embrace AI-powered development tools and infrastructure.
Key Takeaways
Stay Low-Level When Building New Technology Abstractions
Malte explains that AI SDK's success came from restraining themselves from building thick abstractions too early in the AI space. (11:24) Unlike web frameworks where patterns are well-established, the AI application space is still emerging and requires flexibility. By staying at a low level, AI SDK didn't need to be rewritten when the industry shifted from chatbots to agents - it simply composed naturally. This approach allowed them to remain adaptable as trends emerged, rather than being locked into premature architectural decisions.
Never Ship Abstractions You Haven't Used Yourself
Vercel's core principle is dogfooding - they never give users an abstraction they haven't battle-tested themselves. (17:25) AI SDK was extracted from v0, and they continuously maintain this feedback loop by migrating their own tools back to use the frameworks they ship. This prevents the common problem where framework builders, who are usually not application builders, create "ivory towers" that don't work in practice. The constant internal usage ensures high hit rates and reproducible success.
Find Agent Sweet Spots by Asking What Employees Hate Most
The key to successful agent deployment is identifying problems that are tedious, repetitive, and boring - but haven't been automated because they require some text processing or mini-judgment calls. (26:55) Malte recommends asking employees "what do you hate most about your job?" because these problems are often easy enough for current agents to handle while having high business impact. Vercel successfully deployed agents for lead qualification, abuse analysis, and data analytics using this approach.
Solve Recall-Precision Problems with Agent Investigation
Traditional alerting systems face a recall-precision dilemma - tune too aggressively and get false alarms, tune conservatively and miss real issues. (21:37) Vercel's DevOps agent solves this by allowing aggressive anomaly detection that triggers agent investigation instead of human pages. The agent has two minutes to analyze observability data, logs, and time series to determine whether to escalate or dismiss, acting as a "coworker with no sleeping problems" in the loop.
Transform Your Company Around Core Competencies in AI
When adapting to AI transformation, companies must build products that feel native to their existing strengths rather than pivoting to entirely different domains. (37:01) Vercel's success with v0 (web development tool) and AI SDK (framework for AI apps) worked because they extended naturally from their framework company identity. This approach maintains credibility with customers while enabling meaningful transformation rather than forcing uncomfortable pivots.
Statistics & Facts
- TypeScript became the biggest language on GitHub as of the day of recording, surpassing Python in usage. (34:31) This represents a significant shift in developer preferences toward strongly-typed JavaScript.
- Vercel offers zero-cost workflows where developers can run compute for infinite amounts of time when code is paused between steps. (05:20) This eliminates traditional serverless cost concerns for long-running processes.
- Vercel's "agent on every desk" program requires companies to commit to building three agents - the first with full Vercel support, the second with guidance, and the third independently. (31:09) This structured approach ensures companies become self-sufficient in agent development.
