Amp’s new business model? Ad-supported AI coding

From Netflix to Spotify, ad-supported tiers have become a staple component of the digital economy, offering product access in exchange for your eyeballs (figuratively speaking). Now, the folks at Sourcegraph, the company behind agentic coding tool Amp, are testing whether that same model can work in software development.

Until now, the monetization behind AI-assisted coding has been fairly predictable: usage-based API billing or subscription-based access, sometimes softened by a restricted free or trial tier. With Amp Free, Sourcegraph is introducing a third avenue — one where developer time, attention, and data helpd subsidize large-scale access to its AI coding smarts.

As with many similar tools out there, Amp pitches itself as a “frontier coding agent” for autonomous reasoning, code editing, and “complex task execution” from the command-line, or via an IDE extension. Amp has hitherto relied on a prepaid system — users bought credits upfront, drew them down based on LLM usage and associated compute costs, and could top up manually or via auto-reload. Enterprise workspaces paid a premium for added security and compliance, including SSO, zero data retention, and audit-grade visibility controls.

With Amp Free, Sourcegraph is adding a new layer to that structure, where ads and the sharing of training data are the order of the day.

Setting up Amp Free

Enabling the new mode is fairly straightforward. Users can switch to Amp Free directly inside the editor, using the mode selector to toggle from Smart to Free — or through the CLI by running `/mode free`. From there, Amp handles setup and confirms that Training Mode is enabled, a requirement for participating in the ad-supported tier.

Alternatively, developers can visit ampcode.com, sign in, and enable Amp Free under Settings → Amp Free → Enable Training.

Once active, ads – from familiar names such as Axiom, Chainguard, Vanta, and WorkOS – appear discreetly at the bottom of the editor and CLI interface. “The first time you’ve ever seen that,” Sourcegraph CEO Quinn Slack joked in the Amp Free launch video.

Slack also emphasized that Sourcegraph will endeavor to keep ads “tasteful,” and continue improving both the free and paid modes in parallel.

Still, questions quickly surfaced about what data Amp Free collects to make the model training possible. Responding on X to an “avid Amp user” concerned that code snippets might be shared with advertisers, Slack clarified that “ads are shown separately,” and never influence Amp’s responses.

“Code snippets aren’t shared with ad partners, but free mode does require you to opt into training mode, which shares thread data for training with us and the free model providers,” he continued.

What you get (and don’t get) with Amp Free

Amp Free runs a mix of open source and pre-release “frontier” models with tighter context windows, while Amp’s paid (and newly-rebranded) “smart” mode continues to offer the full autonomous reasoning stack.

One key difference: in free mode, users won’t always know which model is handling their request. Some providers, Sourcegraph notes, may “obfuscate” their model names, and the company itself can’t guarantee a specific model will be used, since Amp Free relies on discounted tokens sourced from multiple providers’ surplus capacity.

Unlike ad-supported entertainment platforms such as Netflix or Spotify, Amp Free comes with practical limits. It’s intended strictly for interactive, human-in-the-loop use — meaning developers can use it directly inside their editor or through the CLI, but not for automated pipelines, background jobs, or programmatic API calls. Sourcegraph also enforces rate limits, which haven’t been publicly detailed but are likely to mirror other freemium AI tools: modest per-session or per-hour caps designed to keep the service responsive and discourage automation-heavy workloads.

Amp Free isn’t available to enterprise workspaces or to any team that has disabled Training Mode, since data sharing is required to power the tier. And to manage demand, Sourcegraph says it may need to limit new signups, giving priority to existing users and paid accounts.

The price is right for AI coding community

Clearly, any level of data sharing is a huge no-no for enterprise setups. But once the dust had settled in the aftermath of Amp’s big news, the sentiment seemed to hover on the positive side in terms of how this could open up AI coding to the masses.

Kaushik Gopal, principal engineer at Instacart, said that Amp Free is “kind of fascinating,” noting that while the idea may well have started as a joke, it could quickly become appealing “once we collectively realize the untenable costs of true agentic coding.”

For developers and smaller teams priced out of per-seat or usage-based plans, he suggested, Sourcegraph’s ad-supported model might be the most realistic path to widespread adoption.

And over on LinkedIn, Tinkeracademy co-founder YJ Soon opined that Amp’s new free mode might be how “we finally bring TUI (terminal UI) agentic coding to students,” referring to the prospect of giving learners hands-on access to AI-assisted development directly inside the command line — without the licensing hurdles of GitHub Copilot for Education or the limited functionality of browser-based tools like Figma Make or Gemini CLI.

However you slice and dice it, Amp Free is an odd fusion of two worlds: Silicon Valley’s obsession with AI agents, and the ad-funded logic of old-school media. But if Amp Free finds its audience, it might turn the economics of AI coding on its head.