CHAPTERS
Why AI browsers could replace Chrome for real work
Aakash brings on Naman Pandey to compare three AI browsers—Perplexity Comet, ChatGPT Atlas, and Dia—head-to-head. They frame the goal as finding which browser best boosts productivity (especially for PMs) and where each one falls short.
- •Three tools compared: Perplexity Comet, ChatGPT Atlas, Dia
- •Focus on real workflows: research, automation, and context across tabs
- •Promise: see strengths, weaknesses, and final ranking/recommendation
High-level positioning: research vs tab-memory vs agentic automation
Naman gives a “300-foot view” of what each browser is best for. The comparison is anchored on a simple mental model: research-first (Comet), context-first (Dia), and action/operations-first (Atlas).
- •Perplexity Comet: best for real-time, research-heavy tasks
- •Dia: strongest at remembering and using context across open tabs
- •ChatGPT Atlas: most “agentic”—best at performing actions inside the browser
Getting started: how to install Comet, Atlas, and Dia
They quickly walk through where to download each browser and the basic install flow on Mac. The setup is positioned as lightweight—download and drag into Applications.
- •Comet: download via perplexity.ai/downloadcomet
- •Atlas: accessed via ChatGPT UI (top-left) then download
- •Dia: diabrowser.com/gettingstarted; drag DMG into Applications
Core superpower demo: turning a pile of tabs into a one‑pager (Nvidia research)
Naman demonstrates a baseline capability all three share: using open-tab context to synthesize information without copy/pasting into an LLM. He shows how the browser identifies the relevant stock-related tabs and generates a succinct one-pager with links/graphics as requested.
- •All three can read and reason over open tabs
- •Automatic grouping: ignores unrelated tabs and uses only relevant ones
- •Ideal for creators/researchers consolidating sources into a draft quickly
ChatGPT Atlas demo: auto-filling job applications from a resume
Atlas is shown as a true browser agent: it can operate web forms field-by-field. Naman uploads a resume and instructs Atlas to complete the application, including generating narrative answers for prompts not explicitly on the resume.
- •Upload resume → prompt: “Fill out this application for me”
- •Agent fills fields in real time (Workday-style pain point)
- •Generates responses for subjective questions (motivation/fit)
ChatGPT Atlas demo: LinkedIn “scraping” for outreach and contact extraction
Naman demonstrates Atlas’s ability to navigate LinkedIn-like flows to identify potential podcast guests and assemble a list with profile links. He explains how agent mode can click into “Contact info” sections and compile emails/phone numbers into a spreadsheet-like output, while noting guardrails around the word “scrape.”
- •Agentic browsing replaces custom scraper bots for many users
- •Atlas can infer missing context by visiting a provided website (podcast site)
- •Can traverse UI clicks (e.g., open contact info) and compile results
- •Guardrails: may refuse “scrape” language unless prompted to use agent mode
Atlas + Gmail integration: mining your inbox for recurring subscriptions
They shift to a privacy-sensitive but powerful workflow: connecting Gmail via a built-in integration rather than manual login steps. Atlas can scan receipts to identify recurring charges and optionally include cancellation/support links, turning an inbox into an expense audit.
- •Built-in Gmail connector (authorization flow)
- •Finds recurring expenses via receipt emails
- •Can output a list and link to customer support/cancellation paths
- •Highlights AI advantage: scanning “1,000 things” vs a human’s “10”
Perplexity Comet demo: shopping and price comparison across the web
Comet is positioned as the strongest research browser, especially for time-sensitive comparison tasks. Naman gives a minimal prompt to find gifts for a 10-year-old and compare prices beyond Amazon; Comet returns options with links and prices from multiple retailers.
- •Minimal prompting to test real capability
- •Cross-site comparison even when starting from an Amazon tab
- •Returns links and identifies cheaper alternatives (e.g., Barnes & Noble)
- •Best fit: real-time research and decision support
Comet + Sheets + extensions: auto-populating a spreadsheet and using price-history tools
Naman shows Comet writing results directly into Google Sheets, then iterating to add links and more precise pricing. They discuss why it can access historical pricing—via extensions like Honey/Capital One Shopping—and note how extensions amplify what an “agentic browser” can do.
- •Comet can populate and update Google Sheets from research output
- •Iterative refinement: add links, exact prices, metadata
- •Historical price tracking comes via connected extensions (e.g., Honey)
- •Observation: operations are token-expensive yet often free/fast enough
Reliability check: do AI browsers hallucinate?
Aakash asks whether these tools hallucinate like general LLMs. Naman reports few hallucinations in practice, emphasizing that links and actions tend to be accurate because the browser is grounded in the actual page state and navigation results.
- •Low observed hallucination in links/actions
- •Grounding effect: browsing context reduces made-up citations
- •They quickly sanity-check prices/links against live pages
Dia demo: context-first workflows (YouTube synthesis) and standout onboarding
Dia is praised for product polish and onboarding experience. Naman demonstrates Dia using two open YouTube videos to draft a script with a strong hook, showing it can infer key themes even when videos aren’t actively playing.
- •Dia wins on aesthetics/onboarding polish
- •Strong at using multi-tab context to generate new writing (e.g., scripts)
- •Can synthesize themes from open YouTube pages/videos
- •Limitation noted: more restrictive for Gmail-style automation vs Atlas
Dia for PMs in Atlassian environments: Jira/GitHub/Loom automation
Naman highlights Dia’s enterprise-leaning advantage: tight integration with Jira and the Atlassian ecosystem. Even without a live demo, he describes workflows like generating Jira tickets from GitHub issues/bugs and turning Loom bug walkthroughs into tickets with transcripts and structured fields.
- •Direct Jira integration makes Dia compelling for Atlassian-heavy orgs
- •Potential flow: scan GitHub bugs → auto-create Jira ticket
- •Loom video → transcript + summarized bug report → Jira ticket
- •Best for PM “ticket operations” if enterprise setup allows installs
Weaknesses and risk tradeoffs: speed, navigation friction, CAPTCHAs, and privacy
They enumerate what not to use each browser for. Atlas can be slow for time-critical tasks; Comet may struggle with long navigational chains or dark patterns; Dia raises the biggest privacy concerns around tab context and potential data exposure, based on reports Naman has seen.
- •Atlas: robust but slower for complex step-by-step operations
- •Comet: weaker for long navigation flows, CAPTCHAs, or dark patterns
- •Dia: concerns about controlling/excluding sensitive tab data
- •General advice: none are “privacy-first” by default; evaluate your risk appetite
Building the use-case mind map: PM workflows, general workflows, and non-use cases
Aakash and Naman collaboratively structure a mental model for when to use an agentic browser vs a standard LLM. They categorize PM use cases (competitive analysis, sentiment research, docs generation, structured note-taking) and general use cases (email mining, shopping, scraping), plus situations where these tools fail (login friction, advanced CAPTCHAs, dark-pattern cancellation flows).
- •PM use cases: competitive analysis, sentiment shifts, structured navigation + notes
- •Docs automation: crawl flows and generate documentation with screenshots
- •General: scraping, shopping comparison, inbox audits
- •Non-use: sensitive logins, complex CAPTCHAs, deep/dark navigation (e.g., cancel subscriptions)
Final rankings: overall winner vs best usability, plus pricing/rate limits
Naman ranks Atlas #1 for overall utility (especially agentic actions like scraping), Comet #2 for research, and Dia #3 for narrower flagship use cases. Separately, he ranks usability/interaction quality highest for Comet due to better clarifying questions, with Dia and Atlas close behind; they note Atlas is free to try and $20/month on Plus, with few rate-limit issues reported.
- •Overall: Atlas #1, Comet #2, Dia #3 (most people)
- •Usability: Comet best at clarifying and aligning with intent
- •Atlas positioned as best starting point for newcomers familiar with ChatGPT
- •Cost notes: Atlas free to try; Plus users report minimal rate limits
Advanced tips: capture the “alpha” by testing new features right when they ship
They close with a meta-strategy: these products evolve quickly, so the best ROI comes from experimenting immediately after feature launches. Naman emphasizes watching docs/news and trying new agents early, because pricing/value and competitive parity may change fast.
- •Monitor release notes and AI news; try features immediately
- •Early windows offer outsized value before pricing/limits change
- •Example: new shopping agent experience appeared rapidly
- •Tools will improve; today’s workflows are the “worst they’ll ever be”
Wrap-up: giveaway, where to find Naman, and final calls to action
Aakash explains a giveaway mechanic and reiterates the value for PMs of adopting AI browsers (and potentially getting IT approval for Atlas). They share where to follow Naman and close with standard podcast/YouTube engagement requests and Aakash’s tool bundle plug.
- •Giveaway: comment rankings + DM on LinkedIn within 7 days
- •Recommendation to product leaders: consider Atlas access beyond ChatGPT alone
- •Naman’s handle: @readysetdopodcast (Instagram)
- •Final CTAs: subscribe/follow/review; check Aakash’s tool bundle
