Windsurf CEO Varun Mohan at VB Transform 2025

When AI-powered tools spread via corporate software stacks, the rapid growth of the AI ​​coding platform Windsurfing becomes a case study of what is happening when developers adopt large -scale agent tools.

In a session with today VB conference Transform 2025,, CEO and co-founder Varun Mohan Explained how the integrated Windsurf development environment (IDE) exceeded a million developers within four months of launch. In particular, the platform now writes more than half of the code committed by its user base.

The conversation, moderate by the CEO of Venturebeat, Matt Marshall, opened its doors with a brief but a point of non-responsibility: Mohan could not comment on the largely reported potential acquisition of Windsurf Openai.

The problem has Attracted attention after a Wall Street Journal report Retaining a confrontation of brewing between Openai and Microsoft on the terms of this agreement and broader tensions in their partnership of several billion dollars. According to the WsjOpenai seeks to acquire windsurfing without giving Microsoft access to his intellectual property – a problem that could reshape the company’s coding landscape of the company.

With this context put aside, the session focused on windsurfing technology, corporate traction and the vision of aging development.

>>See all of our Transform 2025 coverage here<

Go beyond the semi-automatic entry

The Windsurf IDE is built around what the company calls a “mental” loop – a draft state shared between humans and AI which allows complete coding flows rather than automatic entry suggestions. With this configuration, the agents can make multi-spinning refuses, write tests of testing and even launch changes in the user interface when a traction request is launched.

Mohan stressed that coding assistance cannot stop at the code generation. “About 20 to 30% of the time of a developer is devoted to code writing. The rest is debit, examines and tests. To really help, an AI system needs to access all these data sources,” he said.

Windsurf recently integrated a browser inside its IDE, allowing agents to test the changes, read the newspapers and interact directly with live interfaces – a bit like a human engineer would do.

Safety and control by design

While AI is starting to actively participate in business development cycles, the accent put by Windsurf on integrated security has proven to be essential. “We use a hybrid model for business deployments – none of the personalized data is stored outside the user tenant. Security is central, in particular with features such as our integrated browser agent, ”said Mohan.

These capacities have already made the board board a viable option for the regulated industries. His agents are deployed on massive bases, including those of Jpmorgan Chase and Morgan Stanley.

Mohan added that, as AI becomes more accessible to the other, security will become a trigger factor for productivity. “If everyone in a business will contribute to technology in one way or another, the missing part is security. You do not want a single system built by a non-technical user destroying another service,” he said.

Small teams, fast tests

Internally, Windsurf is organized in lean squads of three or four engineers, each focused on the test of a narrow set of product hypotheses.

“There is a conviction that companies of a billion dollars in one person are the future. But in reality, more people allow you to develop more quickly and build better products. The key is to organize in small targeted teams that test the hypotheses in parallel,” said Mohan.

This model helped the company quickly in a space where fundamental AI models – and user needs – evolve at dizzying speed.

Large -scale customization

The largest optimization of Windsurfing on a company scale is not a faster generation of tokens or smaller models – it is relevance. “On a scale, the greatest optimization is personalization. Understanding the code base deeply allows the agent to modify large -scale changes that reflect the intention of users, ”said Mohan.

Rather than relying solely on the generation of code for general use, the Windsurf system learns the structure, style and preferences of each customer.

Build for future models

In the meantime, Windsurf designs its platform to remain adaptable as the capacities of underlying AI models continue to grow.

“Each improvement of step passages in foundation models requires a major product to rethink. As agents become more capable, our work consists in building the platform to manage and orchestrate many of them, ”said Mohan.

The company works on an open protocol that will allow companies to integrate any LLM, including on -site models – as part of the Winsurf agent, preserving flexibility and minimizing supplier locking.

Prouvance and measurement value

Windsurf provides transparency in its performance with integrated analyzes. “We provide transparency on the return on investment through measures – such as the percentage of code written by the assistant – which can be linked directly to internal engineering performance,” said Mohan.

This approach allows platform teams to connect agent productivity to the commercial impact, helping to justify new investments.

Targeted execution on flash

Finally, when asked how Windsurf plans to differentiate itself in the middle of OpenAi, Microsoft, Google and others, Mohan has focused on internal speed. “The challenge is not which is more visible today, but which performs the right strategy fairly quickly. The risk is evolving too slowly or projects itself too far and lacks short-term relevance,” he said.

Mohan also rejected the idea that holders are inevitably condemned. “There is no fundamental reason why societies inherited like Salesforce cannot become a native. The real limiter is their speed of innovation, not their capacity. ”

Whether the Windsurfing is part of the future of Openai or continues independently, the traction of the company with corporate customers – and its insistence on the landing of AI in secure and measurable workflows – in fact a player who is worth observed as agent development enters the dominant current.