Perplexity Comet: the browser becomes an agent
Launch & Core Features – On 9 July, Perplexity introduced “Comet”, a Chromium‑based browser that bundles search engine, chat assistant and task automation in a single interface. Tabs are managed by a context‑aware assistant that can make bookings, send e‑mails or fill shopping carts without the user touching any forms.
Early Reviews – Initial tests show that Comet already reliably automates routine tasks (finding restaurants, ordering coffee, adding appointments) while logging every step transparently—a crucial trust anchor.
Roadmap – According to CEO Aravind Srinivas, Comet will roll out “shortcuts for recurring tasks” next week; in future, users will even be able to build “Tampermonkey‑like scripts” in natural language.
Funding – Fresh capital of $100 million raises the company valuation to $18 billion and gives Comet staying power in the race for the AI‑browser crown.
Challenges
Privacy vs. Personalization – A complete browsing history is gold for agents but dynamite for data protection.
Hallucinations & Controllability – When agents act (buy, book, deploy), every step must be traceable; Comet and Edge therefore log actions and offer one‑click reversibility.
Ecosystem Lock‑in – Proprietary “skills” could create new platform silos; open protocols are needed.
What is Chromium?
Chromium is the free, open‑source foundation from which Google Chrome, Microsoft Edge, Brave, Vivaldi—and now Perplexity Comet—are built. It provides the Blink rendering engine, the V8 JavaScript engine and a vast extension ecosystem. Developers thus save years of core‑browser work and can focus on features such as AI assistants.
What is Tampermonkey?
Tampermonkey is a browser extension that manages so‑called userscripts—small JavaScript snippets used to modify websites, auto‑fill forms or overlay extra functions. In the AI‑agent world, Tampermonkey is considered an inspiration for “natural‑language scripts” that will soon handle routine tasks with no code at all.
Current State (Summer 2025)
Perplexity Comet – the agent browser
Launch: 9 July 2025 (Early Access for Perplexity Max)
What it can do: AI search, sidebar chat, tab summaries and—uniquely—complete actions such as unsubscribing from newsletters or placing Amazon orders. Every mouse movement made by the agent is visualised step by step so users can see exactly what happens.
Microsoft Edge Copilot Mode
Since 28 July, Edge has been testing an opt‑in mode in which Copilot can see all tabs, be voice‑controlled and—soon—use history and credentials to complete reservations and purchases.
Coding agents
Cognition has acquired IDE specialist Windsurf to pair its coding agent Devin with a fully integrated development environment.
Open Source
AutoGPT v0.6.16/17 introduces a block SDK, Gmail thread blocks and Excel import—moving the community from experiments toward productive workflows.
Near‑Term Future (2025–2027)
Platform Officially announced Benefit
Comet Shortcuts for recurring tasks & natural‑language scripts “à la Tampermonkey” Finally macros without code; power users share recipes via a marketplace.
Edge Access to history + login data for fully automated bookings Competition for traditional travel or shopping portals; Microsoft hints at a subscription model.
Devin + Windsurf Joint roadmap “IDE‑agent fusion” From issue ticket to pull request—everything in a single agent pipeline.
AutoGPT Multi‑agent orchestration & persistent long‑term storage (roadmap) Open alternative to proprietary agent stacks.
Distant Future (2028–2030 and beyond)
Personal data safes
Local, encrypted vaults that agents access via a standard API—privacy without loss of functionality.
Agent orchestrators
Meta‑agents that benchmark specialised browser, coding or research agents in real time and route them according to reliability.
Energy‑budget awareness
Every task receives a CO₂ or energy budget; agents dynamically optimise model calls.
AGI‑ready UI layers
Multimodal panels (voice, AR overlays) replace the classic chat window; the agent accompanies the user in real time through the physical world.
Micropayment marketplaces for mini‑agents
Developers monetise skill snippets via a Lightning‑like system—app‑store economics for workflows.
Summary
Browsers are becoming the command centre for autonomous agents. Perplexity Comet already shows how AI directly replaces clicks and forms, while Microsoft Edge fuels the mass ecosystem with Copilot Mode. Open‑source projects such as AutoGPT ensure that innovation does not happen only in silos. Transparency—visible agent steps—and the possibility to build one’s own scripts or skills without hitting proprietary walls remain critical for adoption.
