A new project lets users embed ads in AI agent prompts and share half the revenue. Kickbacks is attracting thousands of daily users and raising questions about the future of ad monetization in AI workflows.
Kickbacks is introducing a new way for AI users to monetize their interactions by embedding ads directly into large language model (LLM) agent prompts. The project, led by Andrew McCalip, offers users a 50% share of ad revenue generated during the execution of AI prompts, creating a direct financial incentive for those integrating the Kickbacks code into their AI workflows.
Currently, Kickbacks supports Anthropic’s Claude and OpenAI’s Codex and Claw Code, with plans to expand to Cursor and other agents. According to McCalip, the platform has seen 25,000 downloads and between 5,000 and 10,000 daily users, with hundreds or even thousands of new users joining each day. Average monthly payouts are around $25 per user, while the top 10% of users are earning $75 or more.
McCalip, who is also a founding engineer at Varda Space Industries, brings an outsider’s perspective to digital advertising. He said that users are often skeptical of ads, but Kickbacks aims to make ad exposure more rewarding by sharing revenue directly. The project is designed to be run by a single person with automation, though McCalip noted that new partners with strong ad-serving capabilities will soon be involved.
Fraud prevention is a key challenge for Kickbacks, as automated systems can attract bot activity. McCalip reported that fraudulent behavior is quickly detected due to the lack of human-like engagement patterns, such as circadian rhythms and natural breaks. He also acknowledged the legal gray area of serving ads within third-party AI products, stating that the code has been reviewed by lawyers and that he is prepared to address any concerns from AI labs like OpenAI or Anthropic if they arise.
Kickbacks is also testing a new feature that allows users to opt into data collection and ad targeting, which could increase CPMs by up to five times. The current system serves blanket-targeted ads anonymously, but the upcoming version would allow advertisers to target based on user prompts and context, provided users agree to share their data. This approach raises new questions about data-driven targeting and privacy in AI environments.
As AI agent feeds become more central to user workflows, McCalip believes the monetization potential is significant, with users spending hours per day interacting with these systems. He said that some advertising partners were surprised by the high engagement metrics and ad impression times reported by Kickbacks.
The rise of agentic advertising has sparked debate about whether new models will avoid the pitfalls of traditional programmatic systems. For a deeper look at the risks of complexity and bias in agentic ad models, see this analysis: how automation in digital ad sales may recreate old challenges.
