The New Application Layer - Malte Ubl, CTO Vercel

Key Concepts

• AI Engineering: The emerging discipline focused on building the application layer for AI agents.
• Agents: Autonomous software entities capable of performing tasks, research, and decision-making.
• Vibe Coding: A development approach utilizing AI-assisted tools (like chat SDKs and sandboxed bash interpreters) to rapidly prototype and build.
• Compressed Research: An agentic pattern where AI automates the research phase of a business process, leaving the final decision to a human.
• SaaS Copocalypse: The trend of companies building custom internal software via agents rather than purchasing off-the-shelf SaaS products.
• Agent-as-a-Service: The shift toward software designed primarily for consumption by AI agents rather than human users.

1. The Evolution of Software Engineering

Malte Ubl argues that AI engineering is the legitimate successor to web development. He posits that we are currently in an economic experiment regarding the elasticity of the software market: as the cost of creating software drops (due to AI), the total volume of software produced will increase.

• Economic Viability: Traditional software development was limited by the cost of hardcoding business logic. Agents make previously "uneconomical" software projects viable, filling the gap in the "all software that should exist" Venn diagram.
• Demand for Engineers: Contrary to fears of obsolescence, Ubl suggests the demand for software engineers is rising because companies are increasingly choosing to "make" rather than "buy."

2. Archetypes of Effective Agents

Ubl identifies four practical, low-hanging fruit categories for agent implementation:

• 24/7 Operations: Automating tasks that are currently restricted by human working hours (e.g., customer support).
• Compressed Research: Agents perform the data gathering/research phase of a workflow, allowing humans to make decisions in minutes rather than hours. Example: Vercel’s sales contact routing and abuse report triage.
• Information Surfacing: Agents aggregate existing, fragmented data (Slack, issue trackers, recordings) to provide actionable insights that were previously buried.
• Eliminating Toil: Automating repetitive, low-value tasks to improve employee job satisfaction. Case Study: Vercel’s in-house support agent achieved a 90% deflection rate, allowing human support staff to focus on complex, high-value issues.

3. The Shift to Agent-First Infrastructure

The relationship between software and its users is changing. At Vercel, over 60% of page views are now generated by AI agents rather than humans.

• API/CLI Priority: Ubl emphasizes that developers should prioritize building robust APIs and CLIs over UIs, as agents interact with software programmatically.
• Security Concerns: The current agent landscape is described as a "security nightmare" reminiscent of 1999. Ubl advocates for a separation of concerns: the "harness" (where the agent runs) must be architecturally separated from the "sandbox" (where the generated code executes).

4. The Future of AI Innovation

Ubl presents two potential futures for the industry:

1. Model Lab Dominance: A world where models remain expensive and proprietary, turning engineers into "forward-deployed" workers for big tech.
2. Commoditization (The Preferred Path): A world where models become cheap commodities. In this scenario, AI engineers retain power by building the stable, innovative application layer on top of these models.

5. Notable Quotes

• "AI engineering is the legitimate successor to web development as a really mainstream discipline of engineering that will shape the next decade of software development."
• "The cheaper it is to make software, the more software we're going to make."
• "Eliminating boring work is a very noble mission that we should all kind of strive to do for the companies that we work for."
• "Europe is the leader in AI engineering innovation... we don't have to work at a model lab to drive AI innovation."

6. Synthesis and Conclusion

The main takeaway is that AI engineers are the architects of the next application layer. By focusing on practical agent patterns—specifically those that automate research and eliminate toil—engineers can create immense business value. Ubl concludes that while Europe may not lead in model development, it is uniquely positioned to lead in AI engineering. The future belongs to those who build the stable, agent-accessible infrastructure that allows these models to be applied effectively across all sectors of the economy.