Skip to content

Malte Ubl

Vercel

  • 1 episode on Chain of Thought
  • 5 topics covered
  • 1 full transcript
  • Sep 2025 first on the show
Malte Ubl

Malte Ubl is the CTO of Vercel. On Chain of Thought he shared Vercel’s playbook for moving AI agents from a quick proof-of-concept to production, covering eval-driven development, debugging chaotic agents with tracing, and why prompt injection calls for tool constraints rather than the usual defenses.

Cover art for Vercel's Playbook for AI Agents: From Vibe Check to Production | Malte Ubl

Latest episode with Malte Ubl · EP 39

Vercel's Playbook for AI Agents: From Vibe Check to Production | Malte Ubl

Malte Ubl, Vercel ·

Show notes & transcript → All listening options →

Episodes

What the conversations cover

These notes summarize the episodes featuring Malte Ubl, including other speakers in panel conversations. Follow a source link to read the discussion in context.

EP 39 Vercel's Playbook for AI Agents: From Vibe Check to Production | Malte Ubl

  • Malte Ubl reframes agents not as something weird but as “software that we always wanted to write” — the daily-task automation that needs a little flexibility and was historically hard to model with exhaustive if-this-then-that blocks. Agents do really well on exactly that kind of software, so suddenly it’s super easy to write. Episode takeaways
  • Vercel built what Malte reluctantly calls a “DevSecOps agent” — it takes anomaly detection from a firewall and goes fishing for what happened. Hand a modern frontier model a few tools to query the data stream and it does a pretty good job from scratch; he says you spend an afternoon and have software that used to be incredibly hard to write. Episode takeaways
  • On why coding agents lead, Malte is precise: software engineering has source code you can validate and run unit tests on, so you can reinforcement-learn on this valuable, open-ended problem in a way that’s harder elsewhere. He treats that as a signal the effectiveness most likely generalizes — not proof it already has. Episode takeaways

In their words

Yeah. And the reason just is that I think we, like in our finding, like we tried actually quite a few things, and they all work. Unreasonably effective. And so I feel it generalizes to multiple categories. Now, will give one caveat, which is that the reason why and obviously, we're talking to like a software development audience who is using AI agents today in substantial fashion.

Malte Ubl Vercel's Playbook for AI Agents: From Vibe Check to Production | Malte Ubl · 6:43

Transcript excerpts; ellipses mark omitted words. Some automatically selected quotes have filler words removed.