Perplexity Is Building Its Own Coding Agent, Teammate, to Challenge Cursor and Claude Code

Perplexity, known mainly for its AI-powered search engine, has quietly been building its own coding agent called Teammate. According to a Business Insider report, the company's engineers have been testing the tool internally since May 2026, with the goal of taking over entire programming projects from humans, not just individual prompts in an editor.

According to an internal announcement obtained by Business Insider, Teammate was designed for long-haul work. The tool is meant not only to generate code snippets on request, but also to independently run projects from start to finish, track down the causes of bugs in production, and keep watch over the health of live services.

Model-agnostic, not tied to one provider

The key difference from competitors like Cursor and Anthropic's Claude Code lies in the architecture. Both of those tools are closely tied to their creators' own models. Teammate is meant to remain independent of any single provider, letting Perplexity switch flexibly between models from different companies depending on the task and the cost.

That strategy fits Perplexity's approach so far, having built its search product from the start as a layer on top of multiple external models rather than training and maintaining only its own. Carrying that business model over to developer tools is meant to give the company a pricing edge and flexibility as newer, cheaper, or more capable models appear on the market.

Pressure on engineers

Perplexity's CTO, Denis Yarats, put it bluntly to his own team. According to Business Insider's report, by the end of the year, and ideally sooner, engineers are expected to stop reading code themselves altogether and rely on Teammate.

Engineers should stop looking at code - Denis Yarats, CTO of Perplexity.

Yarats also addressed a common complaint about AI-generated code, so-called slop, low-quality code that later has to be fixed by hand. In his view, the problem disappears if the generated code passes rigorous quality tests, regardless of who or what wrote it.

Slop won't be a problem - Denis Yarats, CTO of Perplexity.

The coding agent race

Perplexity's entry into the fight for the coding agent market comes as competition in the segment keeps intensifying. Cursor, Anthropic's Claude Code, and OpenAI's tools have been competing for developers for months, and every other tech company with its own model or a large budget is trying to break into the same market. With its $20 billion valuation, Perplexity has the resources to compete without rushing a public launch.

For Polish development teams and companies already using Cursor, Claude Code, or Copilot, the arrival of another serious player could mean greater price pressure in the developer tools subscription market and faster development of autonomous agent features that handle whole tasks rather than just suggesting lines of code. It's also worth watching whether Perplexity's model-agnostic approach becomes the new standard, instead of tools staying locked to a single model provider.

Perplexity declined to comment officially on Teammate, so pricing details, an exact launch timeline, and safeguards against errors or misuse remain unknown. Until the company officially announces the product, all information comes from internal documents and people familiar with the matter.

Sources: Perplexity is quietly building an AI coding tool (thenextweb.com), Perplexity is building a secret weapon to join the AI coding wars (aol.com)