Kickbacks: Advertising in Code Agent Loading Spinners

While a code agent processes a long task—compiling, running tests, calling sub-agents—there are seconds or minutes of visible waiting. That idle time, until now, has been worthless to everyone. A project called Kickbacks, shared this week on Twitter by Andrew McCalip and featured on Hacker News, proposes to do exactly that: convert those spinner moments into ad inventory for tools, services, or libraries relevant to developers.

The proposal is conceptually simple: a marketplace where advertisers bid to display short messages during the wait cycle of agents like Claude Code. The developer sees the usual spinner, but with a sponsored message inserted. In exchange, they could receive a share of the generated revenue—hence the name, kickbacks.

What it solves and what it complicates

The timing is no accident. Claude Code has matured enough in recent months that its execution cycles—with hooks, sub-agents, and chained MCP servers—can last several minutes on real tasks. That CPU idle time from the user's perspective is, in advertising terms, a guaranteed impression with moderate attention.

For a development tool provider—a CI/CD service, a niche MCP server, an observability platform—the audience is exactly who they want: active developers, with clear technical context, at the moment they're executing code. The segmentation is implicit.

The obvious problem is trust. Code agents make decisions—which library to install, which API to call, which pattern to follow—and any mechanism of economic incentive operating near those decisions triggers reasonable alarm bells. If the agent or its plugins receive kickbacks for recommending tools, how neutral are those recommendations really? The separation between "ad in the spinner" and "influence on the agent's suggestion" may be technically clear in design, but hard to verify from outside.

Why the MCP ecosystem makes this more interesting (and more delicate)

Claude Code's plugin and MCP server model creates an ecosystem where third parties distribute functional packages that the agent invokes. That ecosystem needs economic sustainability: maintainers of free MCP servers cannot live on pull requests and gratitude indefinitely. In that sense, Kickbacks touches a real problem: how to fund the tool layer of an agent ecosystem without relying solely on subscriptions or Anthropic officially integrating each tool.

However, precisely because MCP servers have access to execution context, files, and network calls, the mix of advertising incentives with that level of access deserves scrutiny. The Hacker News community, so far, has not generated substantial debate in the thread (0 comments at indexing time), which may indicate the proposal hasn't found critical mass yet or simply arrived at the wrong time.

Who this matters to right now

• Maintainers of MCP servers and Claude Code plugins seeking monetization models beyond donations or freemium.
• Engineering teams deploying Claude Code internally who want to understand what new attack surface plugins with monetization logic could introduce.
• Marketplace policy designers within Anthropic, should they consider regulating what distributed plugins can and cannot do.

The idea itself isn't new—screensavers and desktop wallpapers had similar moments—but applied to autonomous agents with execution capability, it gains a dimension beyond simple brand visibility.

---

Editor's note: This is a legitimate exploration of the sustainability problem in open tool ecosystems, but the incentive design needs to be explicit and auditable from day one. Without that, the concept has more potential to erode trust than to generate revenue.