Is the return on investment for generative AI far exceeding expectations? After surveying 1,900 enterprises and IT professionals worldwide, Snowflake found that the average ROI reached as high as 41%! Complete report
The rise of AI Copilot and autonomous agents is redefining the meaning of developers. In a major keynote at BUILD 2025, Vercel CEO and Next.js founder Guillermo Rauch delved into how AI is changing the developer experience. AI is transforming our role from writing code to intentionally guiding intelligent systems. In this future, AI-native tools will significantly enhance developers' productivity, creativity, and scale.
Drawing from his own experiences, Guillermo Rauch systematically dismantled the structural changes occurring in development paradigms under the AI wave. This is not a demonstration about tools or techniques, but a survival guide for contemporary developers: when technical skills are no longer scarce, what is the core capability that determines long-term value? His judgment is clear; we are undergoing a transformation from &#34;pages&#34; to &#34;Agents,&#34; whose depth and impact are comparable to the birth of the first generation of the internet.
<h2>From &#34;Cloud of Pages&#34; to &#34;Cloud of Agents&#34;</h2>
Rauch reflected on the starting point of Vercel. In 2014, when developers still struggled repeatedly with routing, compilers, clusters, and deployment details, only a few large companies with internal infrastructure could efficiently build and deliver products.
The birth of Vercel and Next.js was essentially a &#34;decentralization of capabilities,&#34; allowing the development efficiency that originally belonged only to leading companies to become a default capability for a broader range of developers.
Today, a similar differentiation is replaying in the AI field.
He pointed out that the cloud infrastructure supporting the first generation of the web was essentially &#34;born for pages&#34;: optimizing loading speeds, relying on CDNs, and revolving around one-time requests and responses. However, in agent-driven application forms, this logic begins to fail. The new application forms are no longer page-centric but are centered around continuously running, long-thinking, multi-step orchestrated agents.
As a result, the goals of the cloud are also changing:
<ul>
<li>
Computing power no longer solely pursues &#34;faster returns,&#34; but must support longer, more complex reasoning;
</li>
<li>
Infrastructure is no longer distributed around pages but is centered around &#34;token flow&#34; and intelligent invocation;
</li>
<li>
Search and reach no longer rely on rankings but are actively embedded into users' work scenarios through Agents.
</li>
</ul>
<img src="https://imageproxy.pbkrs.com/https://www.infoq.cn/news/null?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://www.infoq.cn/news/null"/> https://static001.geekbang.org/infoq/b6/b601d563855792e2f91404ecde1a862d.png<img src="https://imageproxy.pbkrs.com/https://www.infoq.cn/news/null?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://www.infoq.cn/news/null"/>
In this sense, he refers to the new form of infrastructure as &#34;Agent-oriented cloud.&#34;
<h2>The interface has not disappeared, it has just become generative</h2>
Alongside the changes in infrastructure, there is a fundamental transformation of the user interface.
Traditional software interfaces are deterministic: developers can precisely control every pixel, anticipating what users will see, click, and which path they will take. However, in the AI era, interfaces are becoming generative and adaptive, generated on demand, changing with context, and highly personalized.
Rauch emphasizes that this does not mean a decrease in the importance of the front end or user experience; on the contrary, experience is more important than ever. What has changed is how experiences are constructed: shifting from pre-designed interfaces to instant generation of the most suitable presentation when needed.
He cites the integration of Snowflake and Vercel's generative UI Agent (V0) as an example: complex data analysis results can generate visual interfaces in real-time during conversations, allowing non-technical users to understand. The underlying trend is the industry's gradual shift from &#34;code-first&#34; to &#34;code-later,&#34; where code is no longer the value itself but a transitional form that serves the results.
<h2>When who can write code is no longer important</h2>
If the changes in infrastructure and interfaces reshape how to build, a deeper change lies in who can participate in building.
Rauch believes that software development is transitioning from a highly specialized skill to a universal capability. In the past, the barrier was syntax, toolchains, and infrastructure; now, the barrier is shifting to a new dimension: whether the intent is clear.
With the help of AI, the ability to express oneself has become a new foundational literacy. Whether one can write code is being replaced by whether one can accurately express what they want. He views this change as an upgraded version of the democratization of JavaScript: if frameworks and platforms once turned many front-end developers into cloud engineers, AI is enabling more people without technical backgrounds to build software for the first time.
In this context, the value ranking of personal skills is being rewritten. Rauch presents a harsh but realistic judgment: do not overly rely on any specific skill. Just as mental arithmetic was once an advantage but ultimately surpassed by machines, programming skills are entering a similar phase.
What truly matters is what he calls &#34;meta-skills.&#34;
<h2>Delivery is becoming the most core meta-ability</h2>
In an era where AI can generate infinite solutions, the real role developers need to assume is not that of code executors, but rather judgers and deliverers.
Rauch views &#34;shipping&#34; as the most critical meta-ability for contemporary developers. It is not equivalent to writing code, but rather an end-to-end comprehensive capability: from problem judgment, product design, implementation methods, to testing, iteration, articulating value, and continuously raising standards.
In his view, AI should not only be a tool for enhancing efficiency but should also serve quality. He repeatedly emphasizes a stance: generating code faster should not become an excuse for lowering delivery standards. On the contrary, true competitiveness lies in simultaneously improving productivity and quality
Around this judgment, he provided specific practical directions: using Agents in customer service, risk control, content generation, data analysis, and other scenarios, allowing the system to take on repetitive tasks and leaving human time for decisions that truly impact product and business direction. As the system continuously learns from feedback of &#34;accepted, modified, rejected,&#34; the quality threshold itself will also rise over time.
Ultimately, he summarized the significance of Agents in one sentence: eliminate the barriers between creativity and reality.
At the end of this sharing, Rauch did not provide a checklist of what tools to learn next, but instead posed a more fundamental question: when the tools are in place and the models are mature, what you really need to think about is what kind of product and value you are prepared to deliver.
The developer experience in the AI era is no longer just about writing less code, but whether you can bring ideas into the real world with higher standards

SNOU

CLOU

SNOW

In the AI era, the developer experience is being redefined. Snowflake's research shows that the average return on investment for generative AI is as high as 41%. Vercel CEO Guillermo Rauch discussed how AI is changing the role of developers in his BUILD 2025 presentation, shifting from writing code to guiding intelligent systems. He emphasized that technical skills are no longer scarce; the core capability lies in adapting to transformation, from "cloud of pages" to "cloud of agents," promoting the widespread efficiency of development