--- title: "Perplexity just launched \"Computer\"." url: https://sam-page.com/thinking/perplexity-just-launched-computer date: 2026-03-02 author: Sam Page type: Post or article categories: [Emerging tech and AI] reading_time: 2 min summary: "19 AI models, 400+ integrations, $200/month. They're betting the future is orchestration rather than one tool doing it all. Here's my take and the shortcomings..." first_published: https://www.linkedin.com/feed/update/urn%3Ali%3Ashare%3A7434276166091374592 --- # Perplexity just launched "Computer". 19 AI models, 400+ integrations, $200/month. They're betting the future is orchestration rather than one tool doing it all. Here's my take and the shortcomings... ![](https://sam-page.com/img/thinking/perplexity-just-launched-computer-li1.webp) Perplexity Computer is a cloud-based system that breaks your task into sub-tasks and routes each on to whichever model it thinks will handle it best. Claude for reasoning, Gemini for research, Grok for speed. One prompt, multiple models working in parallel. It can research, build, deploy and run for days (or weeks!) autonomously. It's a great concept. No single model is best at everything. However there are shortcomings. When you use a model directly (Claude, ChatGPT, Gemini) you control the direct prompt, the context, the tone and the iteration. With Perplexity Computer it wi decide what goes where and stitch them together. That means: - The context window is limited and gets lost - You lost control over the detail and nuance - 19 models sounds smart, but you can get an average output still - Your data is being routed to multiple providers (worth thinking about) - $200/month is a high price tag for tools you can access directly for less In my view, the vision is great. Orchestration is where it's all heading. But it's the difficult bit, and early signs are it's not as good as humans at this still. Perplexity even cancelled a press demo hours before launch because of the bugs. Which probably tells you all you need to know. For me, it's one to watch, but I wouldn't ecourage anyone to rush out and sign up. For now, you're much more effective using the different models yourself and acting as the orchestration layer... Has anyone else tried it yet?