Perplexity brings Deep Research to Computer

Doing deep research with AI has always had an awkward middle step: you run a massive query, get a wall of text, and then copy-paste it into another tool to turn it into an actual presentation or document.

Perplexity just fixed that bottleneck. The company integrated its Deep Research capability directly into Computer, letting users go straight from complex questions to finished deliverables like PDFs, dashboards, decks, and web pages in one run.

Search as Code

Perplexity noted that nearly 26% of all Computer tasks involve research and analysis, making it the single largest use case ahead of coding.

To make this work at scale, Perplexity rolled out an architecture called Search as Code. Instead of querying a search engine step by step, the model treats search as an executable program.

It writes custom code inside a secure sandbox, firing off thousands of parallel retrieval steps for a single question. Because the model watches the code execute in real time, it can deduplicate data, filter out noise, or change course entirely if initial results look weak.

Complex tasks are broken down and routed across more than 20 frontier models. Beyond scraping the open web, Computer connects to internal files through App Connectors and taps premium data sources like Statista, PitchBook, and CB Insights. Perplexity claims benchmark gains on factual accuracy and citation quality across Humanity's Last Exam, BrowseComp, and DeepSearchQA, with every claim still linking back to live source URLs.

Why it matters

We think this is one of the smartest product merges Perplexity has made yet.

Research has never been an end goal for knowledge workers. Nobody conducts due diligence just to read text in a chat window; they need a pitch deck, an executive memo, or a spreadsheet to show their team. By closing the gap between gathering data and generating production-ready assets, Perplexity cuts out tedious workflow friction.

Tapping directly into PitchBook and internal enterprise data also pushes the tool beyond generic web scraping into territory traditionally handled by junior associates.

The clear trade-off is latency and compute. Orchestrating thousands of programmatic searches across 20 models will inevitably take time and burn tokens. But for high-stakes business questions where accuracy and presentation matter, giving up a few minutes for a finished slide deck is a trade most professionals will gladly make.