The New Application Layer - Malte Ubl, CTO Vercel

Malte Ubl, CTO of Vercel, delivered the opening keynote at the first-ever AI Engineer Europe conference, making the case that AI engineering is the legitimate successor to web development as the mainstream discipline that will define the next decade of software. Drawing on Vercel's own operational data and product experience, he argued that both the disruption in how software is built (with AI) and what is built (AI and agents) creates an unprecedented opportunity rather than a threat for engineers.

Ubl opened by addressing engineering job anxiety directly. His thesis is that agents represent a new category of software that makes previously uneconomical automation economically viable. Using a Venn diagram framing, he argued that the circle of 'all software that should exist' is now being filled in — work that was too expensive to build with traditional methods can now be built with agents at modest effort. He connected this to what Silicon Valley is calling the 'SaaS copocalypse,' where companies increasingly choose to build rather than buy software. His conclusion: this elasticity in software demand means more work for engineers, not less.

Ubl outlined four practical archetypes of agents delivering real business value today, rather than speculative future scenarios. First, '24/7 agent-as-a-service' replacements for tasks constrained by human working hours — analogous to existing support agent teams (citing companies like Intercom and Decagon). Second, 'compressed research' agents that handle the research phase of a human decision loop without eliminating the human decision itself: at Vercel, an agent processes inbound contact-sales messages, routing roughly 75% to support and enriching the remaining 25% with LinkedIn and company research before a human takes over — reducing a 15-minute task to near-zero per instance. An abuse-report triage agent follows the same pattern. Third, surfacing latent information that already exists in a company — for example, using agents to keep issue trackers current by synthesizing Slack messages and meeting recordings. Fourth, eliminating toil that employees hate: Vercel's internal support agent achieves a 90% deflection rate and measurably improved support team job satisfaction by routing only genuinely complex cases to humans.

Ubl also highlighted a structural shift at Vercel: over the past 7 days, more than 60% of page views on vercel.com came from AI agents — a data point the company had not previously shared publicly. This signals that software must now be designed for agent consumption, not just human users, shifting the priority from UI polish to robust CLIs and APIs.

On infrastructure architecture, Ubl criticized most current agent harnesses for combining the harness execution environment with the environment where generated code runs, calling it a fundamental security flaw. He noted that Anthropic's new agent product had just introduced the correct separation the prior day, validating his view. He also warned that the industry is entering a security moment analogous to 1999 — a coming 'rude awakening' that demands openness to paradigm shifts.

Ubl closed with a geopolitical and economic framing. He sees two possible futures: one where major AI labs win and commoditize everything, leaving engineers as 'forward-deployed' workers for OpenAI, Anthropic, or Google; and one — which he considers more likely — where model costs commoditize and the application layer captures the value. He cited Vercel's own AI SDK (now serving over 10 million requests per week, led by Lars Grammel in Berlin), the coding agent Pear AI (built in Austria), and OpenClaw as evidence that Europe is already taking a leadership role in AI engineering innovation, even without competing on model infrastructure. His closing message: in the commoditized-model future, AI engineers and the application layer are where real innovation happens.