Ajit Varma on the Future of Firefox and the Open Web

In this follow-up to yesterday’s Premium feature, Mozilla Has a Plan to Save the Internet from Big Tech ⭐️, I spoke with Mozilla’s Head of Firefox Ajit Varma about the future of the open web in the age of AI.

Varma formerly worked at WhatsApp, Google, Uber, and Square, and you can follow him on X/Twitter.

Browsers matter again

Thanks to Netscape and its Navigator web browser, web technologies drove innovation in the late 1990s and into the mid-2010s. So it wasn’t until the rise of the iPhone and Android-based smartphones that the emphasis shifted from the web on desktop computers to app stores and mobile apps on mobile. There was good and bad to that change, in the latter case the sudden rise of junk app store fees and seemingly arbitrary limitations imposed on mobile browser makers.

But with the rise of AI, web browsers, desktop web browsers, are driving another wave of innovation. Some of this is due to Big Tech monopolies like Google and Microsoft that are pushing to deeply integrate AI across their web browsers and other platforms. And some of it is coming from a new generation of smaller companies seeking to disrupt the status quo. But whatever the reason, web browsers are in the news again. And that, Ajit Varma tells me, is good for Firefox.

“One of the challenges we had was that people weren’t talking about web browsers anymore,” he said. “But it’s been an interesting year and a half. We’re having a moment, with new transformative technologies after years of static stability.”

Intent matters

Varma cautioned that while the new web browsers coming from Big AI companies like OpenAI are generating headlines, users need to understand why these applications exist.

“AI companies have an ulterior goal,” he said, “and that goal is to push their AI. But we’re not building an AI company [at Mozilla]. We are looking at the benefits that AI can provide to users, and at providing those tools to those who want it. We’re also providing controls for those who do not want it.”

There, Varma is talking about the so-called “AI kill switch” that new Mozilla CEO Anthony Enzor-Demeo promised the organization would add to Firefox. That update arrived in Firefox 148 in late February, though it was sadly given the more pedestrian name “AI controls” in Firefox settings.

“I was really pushing to keep that name,” Varma told me, laughing. “I wanted there to be an animation” that accompanied the AI kill switch being toggled on. But lawyers and marketing intervened. “The spirit is still there,” he added.

The problem with AI

Even before the current era, AI was being used to describe more and more technologies. Spell checking and grammar checking became AI after the fact, expanding the term beyond the Machine Learning (ML) language that was once more common. The issue, Varma told me, is motivation. And morals.

“AI is an overloaded term, and there is a lot of negativity around the societal impacts of AI now,” Varma said. “But we have to navigate the messaging. For example, is language translation AI?”

For a lot of Firefox users, that’s a line of sorts. They may want to disable most of the AI capabilities in the browser, but language translation is a notable exception, a feature many users find to be quite useful and miss when it’s gone. And so the AI controls interface in Firefox was designed so that translation is at the top of the list of features one can toggle back on after blocking AI enhancements across the board.

“We don’t do dark patterns,” Varma told me. “We talked to customers and ran usability tests, and we want to be prescriptive and make it optional. You can turn it all off. We will never prompt you or harass you.”

Some day, perhaps, the industry will converge on new terminology for AI, Varma suggested, echoing comments I had made earlier about how AI wouldn’t be received so badly if we just referred to it as technology or a similarly non-controversial term.

“We need new words for this,” he said.

Why do some people hate AI so much?

The problem as Varma sees it is that some AI tools are taking away our agency and humanity. The ability for AI to fix spelling and grammar issues is one thing, but then it began suggesting entire sentences, and now it’s just writing for us.

“We see this problem in the MDN Web Docs,” Varma told me, referring to Mozilla’s well-regarded web developer documentation site. “The documentation doesn’t focus on a single [programming] language, but AI coding tools all seem to think that Rust is the best language, always, and it’s creating an echo chamber effect.”

What people crave now, he said, are human-made products and services. This also echos something I’ve discussed, that this AI age may lead to a new wave of so-called artisanal offerings that are made by real people and not AI.

How Firefox is handling AI

Mozilla seems to have landed on a good solution–one might say the correct solution—for handling AI in Firefox. It is simply putting the choice in its users’s hands and letting them decide using controls built into the browser. Today, that takes the form of the AI controls interface in Firefox settings. But in the near future–probably this summer, Varma said–it will expand to include the new Smart Window functionality that Mozilla originally announced as AI Window back in November.

Smart Window is smart because it echoes the Private Window, a well-understood user interface that already exists in Firefox (and, with a slightly different name, in other web browsers). Where you can browse the web semi-anonymously using a Private Window that won’t retain your activities, you will soon be able to use a Smart Window to access the AI functionality of your choice without giving that AI access to your browser context, meaning your open tabs and other personal data.

“Smart Window targets AI explorers,” Varma told me. “It’s a one-way door to AI and an easily-understood concept. There are strong browser controls in Smart Window, and some new user experience features, and you can use natural language to control the browser. You can also pull that work back into a normal browser window. Some people will really like it, and and over time, we’ll look at the features that resonate the most with users and move forward.”

What about local AI?

The AI functionality in Firefox today appears–to me–to be based on cloud-based AI solutions, so I was curious where Ajit landed when it came to local, on-device AI powered by NPUs and GPUs. Oops. As it turns out, Firefox already does use local models: When you enable language translation, that functionality can work offline by using a model that’s only downloaded when you select a language. Other features, like image alt text in the PDF viewer, tab group suggestions, and key points in link previews, likewise involve a model download when enabled.

“We always prefer the most private options,” he said, “but this can be challenging, especially for performance. And online models have more up-to-data.”

In the future, Firefox may add the ability for users to integrate with local AI models similarly to how it can today integrate ChatGPT, Copilot, Gemini, and other cloud-based AI chatbots into the sidebar. “The choice is more important than whether people even use it,” Varma added.

Why Gecko matters more than ever

I have made the case in the past that Mozilla should not waste its time and resources on Gecko, that perhaps the web rendering engine is not the right place to innovate. Varma disagrees, of course. And while that’s not surprising, I am pleased to report that he makes some compelling arguments for continuing this work.

“Gecko is a strength for Firefox,” he asserted. “We are pushing the open Internet forward, pushing for standards, all open source, it’s important. And what if there were just one gatekeeper to the Internet? We feel like having three rendering engines [those in Chrome/Chromium, Safari, and Firefox] is good for the industry, and only one of them [Firefox] is independent.”

“Again, you have to look at the motivation,” he said. “These other companies [Apple and Google] are motivated by profit and shareholder value, they extract value just like app stores do, controlling access.”

One of the issues facing the open web today is whether website makers should build capabilities in for AI agents. The need for open standards and choice is more important than ever in this era, he said, and Gecko is an advantage.

“We have open access to LLMs today,” Varma told me. But this isn’t the case on mobile where there’s no way to publish an app without going through a locked-down app store that especially disadvantages third-party mobile web browsers like Firefox. “The web is the solution” to app stores, he added.

And then there’s also the general issue with most Chromium-based web browsers to consider: They all look the same, and AI is not a choice [in Chrome and Microsoft Edge] but rather an integration forced on users.

More on the problem with mobile platforms

Because Varma had mentioned mobile app stores, I asked him how the locked down nature of these stores had harmed Firefox and whether he felt that the growing antitrust cases against both Apple and Google would result in meaningful changes for his company. (Mozilla spoke out against the enshittification of smartphones, where Apple and Google engaged in the same “platform tilt” that Microsoft did during the PC era, back in 2024.)

“App stores will open up,” he said, “It will happen. When it happens, I can’t say or suggest. But we’ve seen how these companies keep mixing up the letter of the law and the spirit of the law [here, he is referring to Apple’s belligerent non-compliance with its legal requirements worldwide]. And there are places, like Japan, where laws are passed but not followed. It’s challenging.”

The issue for Firefox on mobile is straightforward, and it all boils down to the insanity of these platforms charging developers for the right to develop apps rather than doing everything they can to attract more developers, as is the case on the desktop. This is a frustration I share.

“Software is meant to be free to develop because of the value created there [for the platform makers],” he said. “The incumbent players are preventing access to their platforms so they can control those platforms and lock out choices that might be free, truly free, and better.”

Using AI to fix bugs and find other issues

If there’s a feel-good story about Firefox and AI that even the doubters should rally around, it’s the recent news that Mozilla has partnered with Anthropic after that company used Claude AI to help it find and fix an astonishing number of bugs in Firefox. This was a perfect storm of Anthropic needing a large enough open source codebase to test this work on and Mozilla being receptive, as an organization, to this kind of help.

“We just want to create the best web browser,” Varma said. “Publicly traded companies [that make browsers, like Apple, Google, and Microsoft] would be worried about this kind of exposure because it might impact their share prices and so on. We were like, no, this is great. We get a better product with better security because of it. More companies should operate this way.”

The next few months will be even better on this front, he added, telling me to look out for further announcements about security and bug fixes thanks to AI.

“This is all about giving people the best choice,” he said. “Competition is good, and people will choose.”