The New Application Layer - Malte Ubl, CTO Vercel
Vercel CTO Malte Ubl argues AI engineers at the application layer will capture the most value as model labs commoditize.
- Over 60% of page views on vercel.com in the past 7 days were AI agents, not humans.
- Vercel’s in-house support agent achieves 90% deflection rate; support team job satisfaction rose sharply after deployment.
- Vercel’s sales-contact flow uses an agent to pre-qualify leads and research company size before routing to humans, replacing ~15 min of manual work per inquiry.
- Most current agent harnesses have the wrong architecture by combining where the harness runs with where generated code runs; Anthropic’s new agent product separates them.
- If model labs commoditize (Malte’s likely scenario), AI engineers and the application layer capture business value, not the labs.
- Vercel’s AI SDK exceeds 10 million downloads per week and is led from Berlin; Pi coding agent is from Austria; Open Claw (OpenClaw) also European — Malte claims Europe leads AI engineering innovation.
- The SaaS-vs-build calculus is shifting: more companies are building custom agents rather than buying SaaS, increasing demand for engineers despite AI-assisted speed gains.
2026-04-20 · Watch on YouTube