Quinn Slack puts ads in a coding agent to make AMP free — and it's already covering 80% of model costs

Summary

Quinn Slack, CEO and cofounder of Sourcegraph, is running an experiment that is already showing commercial traction: embedding ads inside AMP, the company's coding agent, to make it free at the point of use. The model is working well enough that ads are currently covering approximately 80% of model costs after the switch to Claude Opus 4.5, which Slack says is cheaper than its predecessor despite being more capable.

The unit economics shifted around during development. Early on, ad revenue covered more than 100% of costs. A model upgrade that made the underlying AI roughly six times more expensive dropped coverage to around 50%. The move to Opus 4.5 recovered much of that ground, pushing the coverage ratio back toward 80%.

How the Ad Format Works

AMP surfaces ads contextually inside the developer environment, combining two valuable signals that most ad platforms cannot replicate simultaneously: sustained screen attention and high-stakes purchase intent. Because coding agents are slow and persistent on-screen, there is natural ad inventory throughout a session. More specifically, when the agent detects a developer's tech stack, it can serve targeted ads from relevant vendors.

Slack gives the example of Cloudflare identifying a user running S3 in a suboptimal configuration and serving an ad that explains the performance benefit of switching, with a one-click action for the agent to execute the migration. That combination of contextual relevance and direct conversion action is difficult to replicate in traditional display or search formats.

The ad platform is already self-serve, and Slack says advertiser demand has been strong, citing high hit rates when approaching developer tool CEOs directly. Competing AI products, by contrast, have largely held back from the format, preferring to let Sourcegraph absorb the reputational risk first.

Strategic Logic

Sourcegraph was founded in 2013 and operates two products: its enterprise-scale code search infrastructure and AMP. Slack frames the ad model as more than a monetization tactic. Free ad-supported usage expands the active user base, which accelerates learning about real-world agent behavior. That feedback loop matters because AMP's stated goal is to remain roughly one year ahead of general-purpose coding tools for the top 1% of developers.

Ad revenue also reduces dependence on enterprise customer commitments that could constrain product direction, giving the team more latitude to swap models and change architecture without commercial penalty. Slack draws an explicit parallel to Google's ad business funding unconstrained internal research.

On model strategy, Sourcegraph chose early on not to train its own foundation model, instead building on Anthropic, OpenAI, and Google. Slack says that bet has paid off. AMP briefly switched its default to Gemini 3 following what Slack describes as a significant quality leap last week, though with Opus 4.5 now available, he expects to switch back. There is no model picker exposed to users.

Slack did not disclose total revenue, user numbers, or valuation figures in the segment. The absence of a paid opt-out tier for ad-free access, which Slack confirmed is not currently offered even on premium tiers, is a notable product decision that could face pushback as the user base grows.