Ajit Varma, Head of Firefox: “As browsers and software become capable of taking actions through AI, the risks extend beyond what employees intentionally share”

Ajit Varma has spent his career considering the role technology plays in the way people navigate, work and effectively communicate across the internet – including leadership positions at Meta, Square and Google. Now, as Head of Firefox at Mozilla, he’s focused on one of the internet’s most notable pieces of software. In this interview, part of our Conversations on AI series, we naturally focus on the role artificial intelligence is playing in the online experience.

There is plenty of hype and even fearmongering, but Ajit doesn’t buy into this. Including the idea that AI will replace the web. Instead, he sees it as a new layer that can change how people interact with the information and services already available. Websites, articles and services all remain intact, but the way we navigate this is shifting.

“What is changing is the distance between someone’s initial intention and the eventual outcome,” Ajit says. While searching once required a single-focus to find a page full of results you may or may not want, AI now can “help organise that information, compare it and act on it”.

The opportunity, Ajit believes, is to make AI feel less like another tool to learn and more like something that can aid with your current task. Naturally, the browser is particularly well placed for that: think of it as a management tool controlling tabs, research and workflows that make up a person’s entire online activity. The browser, he argues, needs to become a “a trusted gateway and control layer for AI,” allowing users a clear point through which they can manage access, protect data and decide which AI services they want to use.

That same principle of user decision making shapes Ajit’s thinking about responsible AI and the growing push toward agent-driven software. While he anticipates agents becoming a greater part of how people use applications, that doesn’t mean the interface disappears. In fact, he believes it becomes more important as AI takes on more responsibility, giving people somewhere to see what is happening and, crucially, to understand what is influencing the experience and offer the chance to step in where necessary. “Choice is a design principle for us at Firefox,” he explains, “every AI feature starts with the same question, how does the user stay in control?”

It’s clear that for Ajit, the question cannot simply be about how much AI can change software, but more about how thoughtfully that change can be introduced and managed by the user-base at large. With that in mind, we wanted to start by asking where he believed AI can genuinely achieve this transformation, and where the industry may be overestimating its impact.

Everyone says AI is transforming software, but where do you believe the industry is still overestimating its impact and where is it underestimating it? 

We’re not seeing AI replace the web today. Instead, it is adding a new layer, helping people explore questions and ideas that no one has yet taken the time to create a website about. It complements how people already use the web.

People are visiting websites, reading articles, comparing products and moving between different services more than ever before. AI can make those activities quicker, but it has not removed the need for the underlying web. Nor does every piece of software need to become AI-first, or every workflow needs to be replaced by an agent. For businesses, the immediate challenge is not getting AI into everything. It is deciding where AI is genuinely useful and where it introduces unnecessary risk or a loss of control.

What is changing is the distance between someone’s initial intention and the eventual outcome. Search once meant typing a question, receiving multiple links and doing the work of navigating between them. Now, AI can help organize that information, compare it and act on it. Ultimately, that shift may prove more significant than any individual chatbot or AI model.

What the industry is underestimating is how urgently the browser must evolve in response. The browser is already where much of our work happens and holds valuable context about someone’s tabs, research and workflows. It is also becoming a new enterprise security surface. Employees can paste proprietary code into a chatbot, upload confidential documents or access AI embedded in everyday software, whether or not IT has formally approved those tools. Increasingly, those interactions happen through the browser.

As browsers and software become capable of taking actions through AI, the risks extend beyond what employees intentionally share. Organizations must also consider what untrusted web content could instruct those systems to do, including through prompt-injection attacks. This changes how businesses need to approach security and data loss prevention. It is no longer enough to secure individual applications or endpoints.

Organizations need real-time, contextual control over interactions such as copying and pasting, uploads, downloads and AI prompts. They should be able to permit certain AI services, block others or disable AI entirely in specific environments. Used responsibly, browser context can make AI far more useful. But it also gives the browser considerable power.

If the same company controls the operating system, browser, search engine and AI model, users and businesses have less visibility into how information is selected, what is surfaced and what, if any, meaningful choices remain available. The browser therefore needs to become a trusted gateway and control layer for AI. It should allow businesses to secure the point where work happens, retain control over their data and infrastructure, and reduce their dependence on a single vendor.

It must also give people the freedom to choose their preferred AI provider rather than steering them toward a homegrown model. Users should be able to understand what AI can access and decide when they want it to be involved. The companies that succeed with AI will not simply be those that deploy it fastest or most broadly. They will be the ones that make it governable by controlling which tools can be used, what data can reach them and what happens to that data. The industry needs to address this now, before today’s design choices become tomorrow’s defaults.

Many SaaS vendors now describe themselves as “AI-powered.” What actually separates companies creating real customer value from those simply adding AI features? 

The real value comes from making an existing experience more efficient and tailored to the person using it. Adding AI to every part of a product does not necessarily improve it.  

At Firefox, we have taken a phased approach. We began with practical features such as summarization, translation, tab organization and optional AI chat integrations – each addressing something people already do in the browser.   

The browser also holds rich context about someone’s activity on the web. Used responsibly, that context can support more personalized experiences. For example, if someone is planning a trip and has multiple tabs open, AI can help identify the relevant pages and organize them into a useful group. This is a clear application of the technology to a recognizable problem.  

However, the use of that context must be based on clear and transparent consent. At all times, the user should understand what data is being used and remain in control of it.  

This is what separates a useful AI feature from one added for the sake of the label. It solves a recognizable problem, fits naturally into the product and remains optional. The aim should be to help users get more done without forcing AI into the rest of their experience simply because the technology is available. 

What has been the hardest challenge in bringing AI capabilities into your products? Technology, data quality, customer trust, regulation, pricing or something else? 

The hardest part has been designing for the wide range of views people have about AI. There is no single definition of AI from an internet user’s perspective, and there is no single preference that works for everyone. 

Some people are enthusiastic about the capabilities of the technology whereas others are concerned about its environmental impacts, job displacement and whether it is ultimately serving their interests. What we found, however, is that people generally want choice and the ability to decide whether AI is part of their experience, if at all, and how it is used.  

Interestingly, only a small minority of users are turning AI features off. What many people value is knowing that controls exist. Even users who never change the settings tell us they are glad to have the option. 

I think of it like a thermostat. You do not constantly adjust it, but you appreciate being able to set it the way you want. The challenge has therefore been less about persuading everyone to feel the same way about AI and more about building a browser that respects those different preferences. For us, that means making AI optional, transparent and under the user’s control. 

There’s growing discussion around AI agents replacing traditional software workflows. Do you see the future as applications becoming collections of intelligent agents, or will conventional interfaces remain central? 

I think agents will increasingly become part of how people use software, but that makes the interface through which people interact with them more important, not less. As agents become more capable, we need to make sure they serve users rather than becoming another gatekeeper. 

We see the browser playing a significant role here because it’s already where SaaS, identity, data and increasingly AI come together. From a more consumer perspective, it can help coordinate agents and automate repetitive tasks, while also giving people a place to understand what is happening, what is influencing the experience, who is acting on their behalf and where they still have a choice. 

We think of ourselves at Firefox as air traffic control. We are not deciding where you fly, but we are clearing the way, helping different airlines work together and giving travellers the freedom to choose and get where they’re going safely. 

The same principle should apply to agents. The goal is not for the browser to make every decision for you, it should be treated as a clear interface between users and the agents acting on their behalf. 

How do you balance innovation with responsible AI? Where do you draw the line between moving quickly and ensuring customers can trust the outputs? 

Choice is a design principle for us at Firefox. Every AI feature starts with the same question, how does the user stay in control? This is exactly why we introduced AI into Firefox in phases and put clear boundaries in place before moving into deeper experiences. 

Responsible AI also depends on transparency. The browser holds rich context about someone’s activity online, and that could support highly personalized experiences. But people need to know when that context is being used, what information an AI feature can access and who is acting on their behalf. Any use of personal context must be based on clear and transparent consent.  

This becomes more important as AI moves beyond finding information and begins organizing it or acting on it. At that point, visibility and clear limits are essential. It should not become an invisible gatekeeper that quietly decides which information, model or service they encounter. 

That is where we draw the line. Moving quickly cannot come at the expense of visibility or consent. If those safe guards are missing, trust will be difficult to maintain. We can continue to experiment and introduce new capabilities within those boundaries. The aim is to make AI genuinely useful while ensuring that people remain able to question the output, make their own decisions and choose how deeply AI becomes part of their browsing experience. 

If you could give one piece of advice to another SaaS executive planning their AI strategy today, what would it be? 

My advice would be to build for user flexibility and be clear about whose interests the AI strategy is serving. There is a temptation for companies with their own models or AI ecosystems to build those services into every part of their product. Executives need to recognize when broader commercial incentives may be shaping product decisions and make sure those incentives do not override what customers actually need. 

Firefox is in a different position as we don’t have a single model that we need to push users towards. Our focus is on building the best browser experience, which allows us to ask whether an AI feature genuinely improves browsing and whether it fits with our core principles of choice, privacy and control.  

That independence helps us shape the product. We can support different providers, use local models where appropriate and allow users to turn AI off. We are not tied to a single model, technology stack or AI ecosystem, which gives us the freedom to make those decisions around what works best for the user. 

For another SaaS executive, I would apply that same test to every product decision. Executives should ask themselves, is a feature being introduced because it makes the product better for the customer, or because it advances the company’s wider AI agenda? 

More interviews

• Rafael Narezzi, CEO and Co-Founder of Centrii: “We need to grow the sector to meet energy demand, but that growth has to come with responsibility”
• Harry Nisbet, General Manager at Marketing Signals: “Shadow AI isn’t just a security issue, it’s a £64m cost problem”
• Dan Clarke, President of IntraEdge Products & Solutions: “More than anything today, AI is fundamentally changing the way we do business”