SUMM

AI tools are enabling a broader population—not just developers—to build software by describing what they want using natural language.
There is a generational leap happening from traditional frameworks to AI-powered coding.
Tools like ChatGPT excel at generating code in popular frameworks like React and Tailwind.
Vercel created Vzero, a text-to-app generator, which has seen explosive growth, now serving over 3 million builders, including Fortune 10 enterprises.
Vzero and similar tools are attracting "dev-adjacent" users such as designers and marketers, expanding the builder ecosystem.

High engagement and retention rates on Vzero demonstrate its value.
Focus has been on fine-tuning models to make generated code reliable and production-ready.
Vercel released frameworks and SDKs to allow others to create AI agents for software generation.
Success is measured with built-in metrics—such as code completion acceptance rates—that guide iterative improvement.
Both objective metrics (functionality, reliability) and subjective qualities (taste, best practices) are enforced in generated apps.
Lessons learned from building interfaces can now be embedded directly into models, making best practices more scalable.

Originally, Vercel aimed to automate cloud infrastructure, focusing on developer experience as a differentiator.
The company's next phase targets automating the software-writing process, not just infrastructure.
Vercel maintains a philosophy of avoiding rigid, strongly held beliefs in a changing AI landscape.
Every aspect of software creation is considered open for innovation or disruption.
The emphasis is on understanding both the developer as a customer and the developer’s end users.

Vzero's growth doubled Vercel's user base year-over-year after launching AI-powered features.
New users generated via AI tools are less focused on technical code details and more on outcomes and user experience.
The paradigm shift positions the agent (AI or LLM) as a key user of APIs, not just humans.
AI-adopted users expect high reliability, putting pressure on product builders for near-flawless performance.

Vercel’s infrastructure evolved from handling static to dynamic sites and is now moving towards generative, streaming content powered by LLMs.
Foundational tech investments (like fluid compute and streaming) are being leveraged for AI workloads.
Emerging protocols (e.g., llm.txt, MCP servers) are supporting agent-driven web interactions.
There are now two webs to consider: one for humans and one for agents, with new protocols facilitating both.

Companies typically start with AI prototypes, move to production AI products, and eventually become fully AI-native across their operations.
Vercel's AI models support both internal functions (like support) and customer-facing products.
Vzero is used to provide expert AI support and even generate code on behalf of users, improving customer service scalability.
Future infrastructure aims for near self-driving capability—handling scaling, troubleshooting, and optimization autonomously.

ChatGPT is now a significant customer acquisition channel for Vercel, as the model frequently recommends the platform.
Vercel’s AI SDK and playground facilitate interaction with multiple LLMs simultaneously, collecting valuable feedback.
LLMs’ preferences are shaped by the data they ingest, leading to organic product recommendations in conversational AI.
Content strategy is shifting: FAQ-style content and LLM-first publishing help match user questions and drive discovery.

The field is moving towards a model where users can describe what they want and have AI generate, deploy, and host apps.
Virtual co-workers (expert agents in various domains) are emerging, and specialized agents will become commonplace in industries like healthcare, law, and finance.
There’s a split between synchronous (instant) and asynchronous (background, collaborative) agents.
Front-end interfaces remain crucial for both output and user engagement; ergonomic entry points for intelligent agents are key.

Vercel distinguishes between users building apps (prefer conversational, low-friction experiences) and those building platforms (require more advanced tooling).
Vzero aims to be as automated as possible, akin to self-driving software creation with minimal human intervention.
Traditional developer tools remain but will likely see declining usage as AI-driven creation matures.

As software creation becomes easier, Vercel emphasizes building better (not just more) apps by embedding security and performance best practices into model outputs.
Vzero's system prevents thousands of potential security vulnerabilities daily before shipping code.
LLMs are increasingly capable of detecting and securing code, such as identifying vulnerabilities in complex systems like the Linux kernel.
Future agents will handle quality assurance and performance optimization, drawing from production data to improve software iteratively.

The traditional concept of permanent, downloadable apps is fading; the generative web will provide just-in-time, personal, ephemeral apps for each user.
Searching for existing software may become obsolete if instant, high-quality generation outpaces the download/install process.
Developers will increasingly build agentic platforms rather than static applications, with agents generating unique experiences for each user.

The next five (even three) years will see massive transformation, with legacy companies being disrupted and AI-native companies rising rapidly.
LLMs’ capacity is underhyped—there’s huge untapped potential even with current technology, enabling new types of engineers and creators.
Everyone will be able to ask questions and learn without friction, shifting how competence is developed.
Vercel’s vision is a generative, agent-powered web, with broad adoption across industries and user types.

Building the Generative Web with AI ft Vercel CEO Guillermo Rauch