Perplexity Quietly Builds AI Coding Agent 'Teammate' to Challenge Cursor and OpenAI

Perplexity, the $20 billion AI search startup, has been secretly developing an internal AI coding tool codenamed 'Teammate' — and may eventually release it publicly, putting the company in direct competition with Cursor, Anthropic, and OpenAI in the fiercely contested AI coding market.

Engineers at the San Francisco company have been using Teammate since May, with tasks already assigned including hunting down bugs in internal systems. According to an internal announcement, the tool is designed for 'long-horizon engineering work' — meaning it's built to own entire software projects end-to-end, investigate issues autonomously, and monitor live services, rather than just autocomplete lines of code.

Notably, Teammate is model-agnostic, meaning it isn't tied to any single underlying AI model — a flexible architecture that could let Perplexity swap in or out frontier models as the competitive landscape shifts. No timeline has been confirmed for a public launch, and Perplexity declined to comment.

Perplexity's CTO Denis Yarats has been pushing engineers to go all-in on AI-assisted coding. Weeks before Teammate's internal debut, Yarats told engineers that by end of year — or sooner — they should effectively stop reading code directly and let AI handle it. He also pushed back on critics who worry AI generates low-quality 'slop' code, arguing that as long as generated code clears quality checks, the concern is moot. The move signals Perplexity's ambition to expand well beyond its search engine roots into the broader developer tools arena.