Code Anything with Perplexity, Here's How

Perplexity is positioned as a faster, more reliable way to build software because it combines multi-agent web search with the ability to switch among different large language models—then feeds that information directly into coding workflows. The practical payoff: when a reference or fix can’t be found quickly with a general chat model, Perplexity can browse and cross-check many sources (the transcript mentions checking 20 or more), which helps reduce dead ends during development.

The walkthrough contrasts Perplexity with ChatGPT by focusing on two differences: model choice and depth of web research. In Perplexity, users can select which model to run (examples mentioned include Claude, Grok, and “gpt-4” style options), while ChatGPT is described as more limited to OpenAI models. More importantly for coding, Perplexity’s web-browsing is framed as deeper and more reference-complete—illustrated with a Bill Gurley example where one model couldn’t locate a specific reference, but Perplexity found it when given the same prompt.

From there, the transcript shifts into a hands-on build: a Chrome extension that tracks time spent on each website. The process starts with upgrading to Perplexity Pro (priced at $20/month in the transcript) to access the “best models,” then selecting a model in settings (the creator favors “Sonnet 3.5”). A key workflow recommendation follows: use two browser tabs. One tab is set to search the web and generate an outline or plan; the other tab is used for writing code without web search, to avoid wasting context.

The coding plan is generated via Perplexity, then executed in a coding environment using Cursor (described as a VS Code fork). The extension is built with a minimal file set—manifest.json plus JavaScript and HTML files (the transcript references creating three files first, then adding popup HTML and background logic). The manifest is configured for a Chrome extension using the latest manifest version, with permissions that allow the extension to store tab-related data and run a background service worker.

Debugging becomes the real story. The extension initially shows tracking numbers in logs, but the popup UI appears blank. The transcript then walks through iterative fixes: reloading the extension, adding console.log statements, and addressing a Chrome content security issue where inline JavaScript in popup HTML is blocked. The fix is to move JavaScript into a separate popup JS file. After that, the UI begins displaying time spent.

Finally, the tracker’s behavior is refined: time should roll up per site rather than splitting across subpages, and the transcript notes that persistence across reloads and correct grouping require additional logic tweaks. By the end, the extension is working well enough to show per-site time totals, and the broader takeaway is that Perplexity’s web search can shorten “error hell” by finding up-to-date fixes and documentation when coding stalls.

Overall, the transcript argues for a workflow where Perplexity handles research and planning (including browsing for fixes), while a dedicated coding assistant (Cursor) handles implementation—yielding faster iteration even when errors occur on the first attempt.