Read time: 6 minutes 36 seconds

Last week I built a tool using Perplexity Computer that turned 45 minutes of unfocused morning Reddit scrolling into 3 minutes of coffee + viral content creation.

The very first post I made using the tool did 200k+ impressions and 500+ engagements on LinkedIn.

I built it in 5 short conversations across 2 evenings - and it made me realise something important about Perplexity’s strategic bet with Computer.

Let’s get into it.

— Tom

Learn how to automate compliance for SOC 2, ISO 27001, and more

Vanta's Agentic Trust Platform helps fast-moving startups and security team get audit-ready fast and stay continuously compliant - turning compliance into a deal accelerator, not a blocker.

Join Vanta's live demo on May 12th to learn how to:

Automate evidence, policies, and remediation across SOC 2, ISO 27001, HIPAA, HITRUST, ISO 42001, and more
Build real security foundations, not check-the-box fixes
Show credibility faster with a public Trust Center and AI-powered questionnaires
Keep engineers focused with developer-native automation

Can't make it live? Register anyway and we'll send the recording.

Save your spot

Own, store, and query your data in one place

PostHog is launching a managed DuckDB warehouse that's fully integrated with your product analytics.

Get direct database access, built-in SQL editor, modelling, and visualisation all-in-one, instead of a fragmented stack.

Pre-register for early access

_{Thank you for supporting our sponsors, who keep this newsletter free.}

How Perplexity wins the 10x knowledge worker

Chess Move

The what: A TLDR explanation of the strategy

Perplexity Computer isn't a better Google.

→ It's the Operating System for high-leverage knowledge work.

Most "AI builders" assume you want to be an engineer. Computer assumes you have a workflow you'd like compounded, and 0 patience for picking a frontend framework or wiring up a VPS.

You describe what you want.

Computer chooses the stack + picks the right model for each subtask + ships you something that runs.

That last part is the fundamental difference between using an AI chatbot and building a functional tool.

Here's what it looks like in practice:

I wanted a tool that scanned 20 tech-related subreddits at 6am for trending visuals, ranked the posts by signal, let me save my favourites, generated 5 hook angles for each save, and pushed them directly into my Notion content queue.

5 short conversations later, that's exactly what I had.

Computer called it Reddit Radar.

The 6am email arrives before I'm awake.

I review the queue over coffee, swipe through 25 posts in 3 minutes, and I'm ready to start my day knowing my social posts are scheduled.

I touched 0 code.

For the audience Perplexity is actually targeting - founders, operators, investors, PMs - Computer is the difference between operator and orchestrator.

This is a piece I’ve been buzzing to share, and I’m pumped to partner with the legends over at Perplexity on putting this together: a showcase of not just the product strategy behind Computer, but also a functional tool that multiplies my own output, to show you (not just tell you) what’s possible on the platform.

Click here to try my tool, Reddit Radar →

Anyone reading this is likely a natural user of Computer, so I’d encourage you to give it a try, and see what you can build when applying it to your own workflows.

Try Perplexity Computer →

💡

Strategy Playbook: Orchestrated systems embed deeper than chatbots.

Breakdown

The how: The strategic playbook boiled down to 3x key takeaways

1. Built for the 10x knowledge worker, not the masses

Reddit Radar is a useless product if you don't know what to do with the output.

It's a feed of trending visuals from technical subreddits with auto-generated hook angles.

To get value from it, you need social media channels to post to, a content publishing engine to feed, and a point of view to apply to a trending visual.

It compounds a workflow that is highly impactful for my business, but redundant for many others.

That's the entire Computer thesis in one sentence.

Perplexity hasn't been shy about this. CEO Aravind Srinivas explicitly told TechCrunch: "We're not on a mission to get as many users as possible." They’d rather impact "GDP-moving decisions" (ie workflows that customers will happily pay for) than maximise casual search queries.

Perplexity has 10s of millions of weekly users; OpenAI has 800 million. They're not playing the same game.

What they ARE playing for: in the few months since Computer launched, Perplexity claims it has performed $2.8 billion in labour-equivalent work for paying subscribers.

More Bloomberg Terminal than Google Search:

Lower usage volume, higher value per query.

2. Orchestration is the moat

When I look at what Reddit Radar actually uses under the hood:

Claude (Sonnet) for the hook generation.
React + Vite + Tailwind for the frontend.
Express + SQLite for the backend.
Notion's API for the content sync.
Gmail API for the morning email.
Perplexity's own cron scheduler.
Reddit RSS parsed in Python.

I pulled the API keys, then I talked to one interface to do the rest.

Never touched any code.

❝

When you build a team, you don't build a homogenous group where everyone has the same skills. You build a team with diverse strengths. We're applying that same logic to AI workflows. The orchestration is the product. The model is a tool.

— Aravind Srinivas

Perplexity isn't trying to ship a frontier model that beats the latest GPT or Claude. Computer orchestrates 19 of them instead.

They're going after the layer that pulls models and maps them to each sub-task - because models converge, but orchestration compounds based on the depth of the workflows embedded within each the customer.

What Computer ships isn't a text block. It’s not a markdown file.

→ It's a system.

The AI hooks Reddit Radar writes are only part of the value.

The rest of the value is in things like:

The WASD-style navigation I built so I could rapidly parse through today’s 25 posts in 3 minutes - keyboard-driven, no mouse, minimal cognitive load.
The sorting algorithm that surfaces highest-potential posts first.
The 6am cron that runs whether I'm at my desk or not.
The Notion sync that pushes each save into my existing content queue with my chosen hooks.
The Sources tab where I toggle, analyse, add, update which subreddits feed the queue.

Each of these features embed the system further into my my business. You can't get any of them from a chatbot alone.

Reddit Radar is a system - a UI, a workflow, a scheduled job, and an orchestration of models touching different stages of a real business process.

That's what Computer ships. Not a smarter chatbot. A built thing, shaped to your business, that you can evolve over time.

Going from chatbot to Computer is like jailbreaking your iPhone in 2009 (iykyk).

3. The opinionated, managed-agent UX is the wedge

Aravind has said he wants Computer to feel "more like using a Macintosh or an iPhone than configuring a server."

The moment I noticed this first hand:

I clicked Save on a post and got a connector error. Screenshot, paste, send back to Computer. Computer diagnosed the issue (apps in the pplx.app sandbox can't reach external APIs, by design), refactored the whole save flow to queue locally and flush via the morning cron, and shipped the fix.

1 message. No code touched. No engineering decisions.

If I'd built this myself, I'd have spent two hours fighting an OAuth flow between Vercel and Notion, and probably never shipped a working prototype.

Computer routed around its own constraint.

This is the explicit design philosophy. Dev-heavy agent stacks like OpenClaw require terminals, server config, and permissions.

Computer calculates and executes on the workaround instead of handing you the limitation.

Pick the highest-value-yet-most-annoying 30 minutes of your week. Describe what you'd rather happen instead. Computer will build it before you finish your coffee.