The Future's Best Ads Won't Be Seen by Humans ⚡

CV Deep Dive

Today, we’re talking with Ryan Hudson, Founder of ZeroClick.

ZeroClick is building the ad platform for AI. Founded by Ryan Hudson - founder of Honey, the browser extension acquired by PayPal for $4 billion - ZeroClick enables brands to integrate contextual advertising directly into AI systems’ reasoning processes. Instead of interrupting users with display ads, ZeroClick’s “reasoning-time” ads enrich AI-generated answers with promoted context from advertisers, creating a new monetization rail for AI developers to be able to sustainably grow while offering broad, generous access to their tools.

The company launched in August 2025 with $55 million in funding from Anthos, Ludlow, Anfa, Wonder, Protagonist, and Hudson himself - the same investor group that backed Honey. ZeroClick quickly followed with the acquisition of Sleek, a YC-backed browser automation platform, signaling a bet that the future of AI advertising lives not just in chatbots, but everywhere users interact with AI - starting with the browser. The platform already connects over 10,000 advertisers with millions of users.

In this conversation, Ryan walks us through the founding story of ZeroClick, why AI’s monetization crisis creates an opening for a new kind of advertising, and what he’s learned from over a decade of building at the intersection of browsers, commerce, and consumer products.

Let’s dive in ⚡️

Read time: 9 mins

Our Chat with Ryan 💬

Ryan - welcome to Cerebral Valley! Take us through your background and your path from Honey to ZeroClick.

I’ve been at it for a while now, longer than I’d like to admit. I started Honey probably close to 14 years ago. It was a great journey - a cool, consumer-facing business where we had a chance to help people save a ton of time and money. Along the way, we built an awesome team. After we sold to PayPal, a bunch of us still liked building things together.

About two years ago, I started talking to some of the people I’d worked with before and we realized we’d like to build things together again. We didn’t know exactly what that might look like or what it would be, but we had a core thesis that the economic engine for the Internet was not quite working well for everybody.

The initial products we launched under the Pie brand were actually an ad blocker that gave users control over their advertising experience. Instead of brute force ad blocking, the idea was that if you gave consumers an incentive to participate willfully in the type of advertising they were interested in, you might be able to help content creators and publishers persist with their advertising model while giving people control over their ad experience.

We built Pie Ad Block and scaled it to over 2 million users. Then about a year ago, we ran into issues with Google’s reinterpretation of their single-purpose policy for Chrome extensions. We had an existential crisis - it seems like everybody in AI is having some version of this every day - of who we were and what we were trying to do. If we weren’t going to be able to build the best ad-blocking experience, we needed to go back to the drawing board. What is the role of an ad blocker in the world of AI experiences? People are shifting more of their attention and economic choices into ChatGPT, and it was clear that AI is this powerful new platform that’s going to change how everything works.

So we started poking at what the monetization of the Internet might look like in an AI-first world. We saw that everything had basically converged toward paid subscriptions as the only monetization model. It felt like that was going to leave a lot of people and a lot of experiences behind. If you were only able to monetize AI with a paid upsell - “here’s the free tier, we’ll rate-limit you super early and try to upsell you into our paid plan” - that works for a lot of experiences and types of buyers. But what if there was a different monetization rail?

If you rewind in history, any large-scale consumer-facing business has some combination of a free tier, an advertising-supported layer, and a paid version. We looked around and asked: who’s building the monetization rail for the free tier of AI? We didn’t see anybody doing it. So that’s how we came to start working on ZeroClick. The basic bet is that advertising can be a monetization primitive - a standalone rail. It’s only an artifact of very good acquisition strategy and execution that we think of advertising as something that sits only inside of the big platforms, whereas payments sit outside of it like Stripe. There’s no reason why the advertising experience also shouldn’t be like that. In cases where the payment rail is inside the platform - like the Apple App Store - the market rake is 30% instead of 3%. Even if the large platforms build their own ad ecosystem, there’s still a huge opportunity for a third-party independent advertising platform.

You built an incredible team at Honey, had a $4 billion exit, and then decided to do it again. What pulled you back in?

Building is core to what I find fun, and building with a team of like-minded, highly capable people is extra fun. I dabbled in investing and came to the conclusion that I either wasn’t good at it or certainly didn’t like it. I had an entrepreneur’s flaw approaching it - if I hear somebody’s idea, most of the time there’s a decent amount of thinking core to it and I’m like, “Yeah, I can see how I can make that work.” That’s not what the role of an investor actually is. Even while doing it, I wanted the building part. It was like confirming that I just wanted to build things, not participate in other ways.

Sitting out and doing nothing during the time post-acquisition - I had to do that for a while. I built other things, physically, like construction projects. They don’t have the same scale. There’s something rewarding about trying to build something that probably isn’t possible and probably isn’t going to work, where it’s like, “You might as well try.”

Break down ZeroClick for us. What does “reasoning-time advertising” mean in practice? Walk us through what happens from the moment a user asks an AI a question.

There are a few different versions of this - admittedly, we’re figuring out the product-market fit. We’re doing a lot of experimentation with different approaches. The first thesis is essentially that AI systems would benefit from a paid search tool that they use alongside their organic search tools. In a world where organic search results get gamed by AI systems even more than they already have been - and there’s no user signal of quality - a paid search environment creates the opportunity for a different type of signal.

Running an ad blocker, we saw that in many high-intent situations, the paid advertising was actually better quality to the user than the organic results. Our ad blocker has a visual mode where we show you the ads disappearing. In Google Shopping searches, people were saying, “Stop making the right answer disappear.” On Amazon, it’s loaded with advertising, but a lot of times it’s well-targeted to what you want. Especially in new product categories, things where the answer changes over time, or cases where the answer to your query has additional context - like an offer or a deal - that paid version can have actually useful information that’s not available otherwise.

So the core insight is: there is signal embedded in advertising that’s useful, not detractive from the experience. In its most integrated version, our advertising system is essentially a paid search RAG tool that the AI system can use to integrate into a response if it determines the information is helpful or relevant.

In many instances, this is super powerful - and it’s the version that people are most afraid of, thinking, “Was my AI unbiased or not?” That’s a fair question. As a result, we have other approaches where the answers are not fully integrated into the LLM response. Developers of different experiences often don’t want that either - they’ve tuned their product for a particular use case and anything that changes it makes them uncomfortable.

One area where we’ve done a lot of experimentation is in dev tools environments. We’ve started partnering with companies that have products for writing code and selecting platforms. Effectively, all the models have their default stack - they’re highly likely to send you towards Vercel with Supabase. There’s an emergence of one vendor per category as the default. But there are a lot of other options that can be more tailored to what you’re actually trying to do. Being able to put those opportunities in front of an LLM in a highly contextually targeted way - it’s not tweaking the code your IDE is generating. It’s saying, “Hey, have you considered this other tool, this new company that wasn’t in the pre-training data because they didn’t exist yet?”

It’s efficient, context-aware matchmaking. The extra layer is that your AI agent system can be the one reviewing this information and deciding if it is or isn’t actually good. The ad load doesn’t even necessarily have to land on a human. Increasingly, agents are making economic decisions on behalf of people. That’s very much central to the name - what if there’s no click? There’s no human doing the deciding, but there’s still enormous economic value being created for an advertiser or platform. That’s not how existing advertising systems are designed. It’s kind of the opposite.

There’s a lot of ambient panic about advertising in AI chatbots. But there seems to be strong incentive alignment for the information being served - even if monetized - to be useful. How do you think about that tension?

The question is: do we want the game to be overt and efficient, or buried in the background polluting all of our other channels? That stuff is probably going to happen anyway, and it’s getting easier to do. My bet is that the commercial signal in an advertising platform becomes a useful tool to agent systems - they actually want this information. Better answers with the information than without it.

From a human cognitive load perspective, you effectively have an agent that watches all the ads for you. You don’t even have to see them. You have something watching ads on your behalf to decide if there’s interesting stuff for you. All of the reasons we built an ad blocker - humans don’t want to be bombarded, it interferes with cognitive load in distracting situations - but now in an agent environment, the agent as an economic decision maker can consider stuff it wouldn’t otherwise.

My wife is doing amazing stuff with OpenAI - like building Instacart orders. She doesn’t have to see all the offers from all the brands. But it creates an entirely new advertising interface where there are probably tons of CPG brands that would love to advertise into an agent-driven Instacart builder. All of a sudden, it’s offers that her agent is evaluating based on what it knows about her. It can do efficient coupon search and all this stuff that, from an economic theory point of view, is insanely value-creating for the user and the brand. There’s this efficiency of market-making, and it’s not wearing on the human attention span.

How does ZeroClick concretely solve the monetization gap for AI developers?

We are monetization rails for AI experiences - usually supporting a free tier, but it doesn’t mean you can’t have a paid tier as well. Advertising is useful. Ads for agents are actually good. We can create different advertising experiences to go along with any developer’s AI traffic, and we have a range of partners spanning developer tools, interesting new software products, and consumer experiences integrating shopping offers. We span a bunch of different categories and can help monetize different experiences with contextual ads for AI systems.

You launched before OpenAI introduced ads into ChatGPT. Perplexity had their Sponsored Questions. How does ZeroClick fit into a world where the platforms themselves start running ads?

It creates more interest and understanding on the advertiser side, which is helpful. People are starting to ask, “What is advertising going to look like in ChatGPT?” I think even with their launch, OpenAI will be pretty cautious in how they roll it out and do a lot of testing with a small number of brands. It’s not available to advertisers broadly today. I suspect it will be at some point, but it’s a way for people to start learning about how to advertise in different AI interfaces.

The areas I can’t see OpenAI moving as fast on are purely agentic flows. They’ve been pretty clear about wanting to be careful about stepping on results in a way that changes the core ChatGPT experience. Their ad platform appears to be leaning more towards Facebook-Instagram-style profiling - using aggregate context accumulated through all your chat sessions to do lookalike modeling. Sam has said Instagram ads are good, and I agree. They actually feel like content. But I don’t think that’s the only answer for what ads to monetize AI experiences can and needs to look like.

Perplexity, by the way, subsequently stopped doing Sponsored Questions and pulled back from advertising. I think they’re intentionally differentiating from ChatGPT, which is probably smart. But for us, it doesn’t really matter what these chatbots or services are doing within their own silos. We’re building the infrastructure - a separate system for the rest of the ecosystem.

For most developers, it doesn’t make sense to have your own direct ad sales team. When the Internet started, in the earliest days, you had your own direct sales team selling to advertisers if you were a website. That only works at certain scale. There’s clearly a role for someone like us to be the outsourced sales team and create market liquidity with enough volume to be interesting to advertisers and meaningful for monetization.

Do you see this as a winner-takes-most market? Are you concerned about moat?

I do think it’s a winner-takes-a-lot type of dynamic. One of the reasons I think this is interesting to work on now is that the cost of software development is dropping dramatically. The output from individual developers or small teams has massively increased. In that world, a lot of software becomes relatively commoditized and replaceable. People are just replicating open source packages instead of importing them.

The really big software businesses going forward will be ones with network effects inherent to the product. It’s either effective scale or network-effective usage - having more users adds to the value. It’s always been this way somewhat, but it gets amplified.

On the other end of the spectrum, I think you’ll see a massive increase in customized, bespoke, low-user-count experiences - so tailored to somebody that they become worth paying for. I could see a resurgence in VR, for example. These platforms didn’t sell enough units to attract AAA-level development investment, so nobody built the killer app for them. All of a sudden, if one person can make a AAA game and doesn’t need it to be Grand Theft Auto for it to be successful, maybe the killer app for these platforms gets created by indie developers or small teams.

I’m pretty bullish on that. In my own side projects, I’ve learned you can build for platforms you know nothing about. If they’re well-documented, you throw Claude Code at it and you can one-shot very complicated software on platforms you’ve never touched before.

Talk us through the Sleek acquisition. You’ve said the future of AI lives in the browser - why is that your bet over standalone AI apps?

The browser is effectively the neutral platform where all the information and interfaces have been built. Absent AI deciding to build its own fully optimized version of the same thing - which I think we’ll get pieces of - until that fully happens and people stop doing any browsing at all, it’s the core neutral platform.

We built what you could argue was the original agentic browser automation capability. Others had done stuff in extensions before, but building a commercially interesting multi-step integration across the whole Internet - tens of thousands of stores that don’t have an API - figuring out how to do that is foundational in our team’s DNA. Now agents are doing that and they’re insanely more powerful, but they’re effectively making the breadth of the Internet composable into different skills.

I can see a return to something akin to the Web 2.0 era of mashups - people taking this thing from here and that thing from there, gluing them together with AI. A lot of the creativity of that era was really fun as a builder. It’d be extra fun to see how much people can do now with those concepts and how sophisticated it feels.

You brought most of the Honey team with you, same investors too. What does that continuity mean for how you’re building ZeroClick, and what are you doing differently this time?

For the initial phase - the first year and a half - it meant insane alignment on how to build. Very cohesive as a team. The thing over the last three months that everyone’s realizing is that sprint planning and doing software development the way we’ve all learned to do it for the past decade is probably wrong now.

I introduced a very liberal expense policy for AI tooling early on. Your budget as an employee is effectively uncapped as long as you’re using it. Get the best version of whatever tool you’re using. Don’t live in the past. You have the opportunity to pay $200 a month to live six weeks or three months in the future - you absolutely should.

This year, as the fully agentic capabilities have emerged, I have a sense that AGI has been achieved inside the big labs - they’re just not likely to release it as a product anytime soon. However, all of the pieces of AGI are available to anyone that wants to assemble them. The models are capable. It’s all harness engineering - putting together the right context, maintaining state and memory access. I suspect about six months ago, internally, they figured it out. That’s why you saw what at the time seemed like insane capex commitments to building massive data centers. The second you hit AGI, you’re like, “I need more compute.” So you go sign $200 billion capex deals.

What was unexpected is that tools like Claude Code kind of showed the broader community that maybe you just need the models already available - with the new hardware capabilities coming out, you can do a lot. We have a project that’s completely isolated, separate infrastructure, that we think agents can operate freely in. I’m really excited to see the experimentation that happens.

You’ve been building in the browser, commerce, and ad space for over a decade. What’s the most counterintuitive thing you’ve learned about how people interact with ads?

Ads don’t have to be bad - that’s the core of it. Inside the Honey experience, advertisers were paying to give you cash back to help drive you towards conversion. That’s a perfectly aligned version of a consumer advertising experience. It fed into this worldview I have: advertising can be win-win for consumers and advertisers. The most value-destructive thing is not having that marketplace of information exchange happen efficiently.

When advertising is loud and noisy and poorly targeted, it’s inefficient and value-destroying. But when it’s well-targeted and interesting, it actually creates a ton of value. It shows you something cool you might want to buy. Efficiently matching you to that opportunity is great for how capitalism and markets work - it’s the information piece of the marketplace coming together efficiently.

Super Bowl advertising - most people think it’s maybe better than average. That’s the product for watching the Super Bowl: the ads. It’s a case where advertising is super valuable to the advertisers reaching a massive audience, and people watching enjoy the value of the ads. I think there are cases like that for most types of advertising. That may not be obvious.

What does ZeroClick look like in three years? What’s the version you’d consider a massive success?

Projecting three months, let alone three years - the asterisk is obvious. I’d like us to be thought of as the monetization rails for AI systems. The way people think of Stripe if they need payment or subscription rails - if we’re successful, that’s how people will think of ZeroClick.

There’s a lot that has to go right between here and there. Most of the time, these things don’t work out. But we’re going to have a lot of fun trying, and hopefully we can support entirely new types of businesses and products that can’t exist without that rail.

In the shorter time frame, hopefully by the next YC class or two, entrepreneurs are starting businesses with the understanding that this monetization rail exists and can be a part of their product pricing strategy. If you’re an indie dev and want to try building with ZeroClick, you can reach out to our team today. We’re doing pretty bespoke onboarding to tailor everything to each developer’s experience, but we can work with anybody at any scale.

Last question - what’s your most contrarian take on where AI advertising is headed?

I think it’s going to be in most places. Including products that have professed they’re not going to do it. Google was the original one - “We’re not going to have ads.” Anthropic says they’re not going to have ads. Perplexity probably comes back around when it’s no longer a differentiator. Other companies will spring up that have advertising as an option, and being able to reach billions of people with a free tier that monetizes is probably going to out-compete something that’s only available to paid subscribers.

Conclusion

Stay up to date on the latest with ZeroClick, follow them here.

Read our past few Deep Dives below:

• OpenAI’s Codex Lets One Engineer Ship Like a Team of Ten

• Smack is Building the First Defense AI Lab for Decision Dominance

• Augment Code is Your AI Coding Agent for Real-World Codebases

• DeepSource - The Vulnerabilities Every AI Tool Missed

• Daytona - Composable Computers for AI Agents

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.

CV • Instagram • X • All Events