Andrew McCalip is a founding engineer at Varda Space Industries, where he oversees the manufacturing of things like hypersonic reentry vehicles and satellite buses (which are the central satellite bodies that carry a payload or instruments, not something that, like, shuttles satellites out to a launch pad).
But he’s always had a taste for online virality, he told AdExchanger. You might not know his name, but perhaps you recall the story from a few years ago of an American who attempted to recreate the South Korean LK-99 semiconductor breakthrough. Or, more recently, the person who built the orbital data center cost calculator.
Another of McCalip’s other self-described “really aggressive hobbies” is his autonomous boat, which, he says, will be able to sail itself around the world using Starlink Wi-Fi.
Those projects have achieved social virality as tech media stories, but the first of McCalip’s ideas to become an actual potential money-earning venture is called Kickbacks, which allows LLM users plug the Kickbacks code into their AI infrastructure and receive ads during those moments or minutes of time when a prompt is being executed by the agent.
Right now, Kickbacks works with Anthropic’s Claude and OpenAI’s Codex and Claw Code, though McCalip said he’s working on adding Cursor and other agents as well.
The reason users want to introduce ads is that they receive a 50% cut of the revenue.
Right now, there are 25,000 total downloads and about 5,000 to 10,000 users on a given day, he said. But hundreds or sometimes thousands of new users are joining per day as well, he added.
The average user is seeing about $25 per month in ad rev-share payouts, he said. The top 10% of Kickbacks users are clearing $75.
“Everyone’s trained to see ads as this parasitic thing,” he said before adding: “Sorry to the entire industry, but it doesn’t have the best optics.”
What’s next?
McCalip brings an outsider’s perspective to his new ad revenue project, which has pros and cons.
Almost never do users or the audience of an ad impression benefit so tangibly by the exposure as they do with Kickbacks. McCalip describes Kickbacks as having a Robin Hood-type appeal, but another way to phrase it is that it’s an outsider’s approach to online ads.
McCalip said he also doesn’t plan to hire an ads industry expert, at least not until he has to. “I really want to see how far I can push this as one person with automation.”
But Kickbacks will soon be bringing in a couple of new partners. McCalip wouldn’t specify what type of company or companies, other than saying he needed the help of firms that could bring “a deep ad book and a great ad-serving engine.”
And so far, McCalip has been able to handle the onslaught of bot fraud.
Any automatable payout system – in this case, running agentic prompts and clicking links to generate ad revenue – will attract fraudsters, he acknowledged. The bots are recognizable, he added, because they have none of the “circadian rhythms” of human activity, with random spikes in engagement but also coherent, linear prompts themselves (which are visible to Kickbacks), and different types of engagement in the morning and at night, as well as clear breaks for meals.
“The fraud pops out super fast,” he said.
Kickbacks may confront an even more insuperable problem than fraud if and when it deals with its own status in the demilitarized zone of AI chatbot and LLM advertising.
McCalip said he had lawyers thoroughly review the code and will “definitely have that conversation if the AI labs ever reach out.”
It is not impossible that OpenAI, Anthropic or another such AI lab will come to view Kickbacks as a breach of terms of service involving ad serving within their products.
“It’s interesting because the labs, they kind of do want this to happen,” he said of Kickbacks’ ad-based business model.
If the LLM served the same prompt-loading ad, for example, it would likely be met with public revulsion, even if the LLM offered a rev-share or a discount on an enterprise subscription. But Kickbacks doing the same thing feels cheeky and fun.
Kickbacks is “absolutely threading the needle of public opinion,” McCalip said.
Eventually, too, there are ways to benefit the LLM within the system. Rather than payouts via Stripe, for example, the revenue could go to a wallet attached to, say, Codex or Claude, and thus cover the costs of a subscription and use of the product itself.
Another new product that’s coming out with the updated version of Kickbacks, which is currently in beta testing, is a way for users to opt into data collection and ad targeting, with the promise of up to 5x greater CPMs from their ad rev-share.
Right now, Kickbacks’ ads are all anonymously blanket-targeted. The new version with enhanced user opt-in would have login-type data (name and email), as well as users agreeing to “share the context and the prompts themselves,” he said. That way, advertisers can target based on the types of prompts and work being done by those users.
Another element of being an ad industry outsider is walking into the minefield of issues that come with data-driven targeting, such as beginning to collect and target based on AI prompts.
The rewards could be great, however, not just for McCalip but for users and their rev-share.
“This will be one of the most highly monetized surface areas in the entire world coming up,” McCalip said of AI search agent feeds.
“The data is incredibly rich,” he said, and people are staring at that window for four to 12 hours per day.
“When some of the advertising partners I’m working with now saw the metrics of how much time and how many ads [are served to Kickbacks users], they didn’t really believe me at first,” he said. “The impression time is just outrageous.”
