The Browser Is Becoming the Creator’s Co-Pilot — and That Changes Your Research Workflow
The most underrated bottleneck in social media work isn’t posting. It’s the 40 minutes before you post: pulling competitor pricing, checking what a rival brand said on their site, grabbing a stat from a PDF, comparing three flight options for a shoot, cross-referencing a claim before you put it in a carousel. That research tax is invisible in analytics dashboards, but it’s where a lot of creator time disappears. So when I saw Peji — an iPhone app from Matteo Ricupero that pairs an AI chat with a real WebKit browser you control — my first thought wasn’t “another chatbot.” It was: this is a research-assistant shape, and research is the unglamorous half of content operations. Here’s what it actually is, where it fits, and where I think it falls short.
What Peji actually solves (and what it doesn’t)
Strip away the launch-page language and Peji is a mobile agent with a browser attached. You ask for something — a price comparison, a flight, a PDF summary — and the app opens real WebKit tabs, reads them, clicks and types, and keeps going until the task is done. Every step shows up in the chat, and the app stops and waits for your tap before it sends, buys, pays, or deletes anything. Chats, files, and dictation stay on the phone. When the answer is a price list or a schedule, it comes back as a table or a chart with source links. It’s free on the App Store, with a Pro tier that “adds usage” — the exact limits and price are not disclosed on the page.
That “stops and waits for your tap” detail is the whole ballgame for me. Anyone who has watched an autonomous agent confidently click the wrong button knows the difference between a demo and a tool you’d actually leave running while you make coffee. A human-in-the-loop confirmation gate before irreversible actions (send, buy, pay, delete) is the minimum viable trust contract for agentic software in 2025. The team is essentially saying: the agent does the boring navigation, you keep the keys to anything that costs money or can’t be undone.
The part creators should care about: source links
The feature I’d actually use daily is the sourced output — tables and charts with links back to where the numbers came from. In my experience, the failure mode of AI research tools isn’t that they’re wrong; it’s that they’re wrong plausibly, and you can’t tell without re-doing the search yourself. If Peji genuinely returns a comparison table where each cell traces to a tab it actually opened, that collapses the verification loop from “re-research everything” to “spot-check three links.” That’s a real workflow change, not a parlor trick. I’d test it by asking for something I already know the answer to — a competitor’s current pricing page — and seeing whether the cited link matches the number.
How it differs from the tools you already pay for
The obvious comparison set is split in two, and Peji sits awkwardly between them.
On one side: ChatGPT, Claude, and Perplexity. These are excellent at synthesis and increasingly good at web search, but they don’t operate a browser on your phone in a way where you watch each click and approve each commitment. Perplexity cites sources; it doesn’t fill in a form for you and pause before submitting. The distinction matters for tasks like “find the cheapest of these three hotels for the shoot dates and hold the one with free cancellation” — that’s navigation plus judgment, not summarization.
On the other side: desktop agent browsers and automation stacks — Comet, Dia, and the various “computer use” agents. These are powerful, but they’re desktop-first and often assume you’re sitting at a keyboard supervising a long session. Peji’s bet is mobile-first and narrow: the phone is where a lot of creators actually do their between-shoots research, and a phone browser agent that respects tap-to-confirm is a different product category from a desktop automation cockpit.
And then there’s the tooling creators already use for scheduling and analytics — Buffer, Later, Metricool, Hootsuite. Peji doesn’t compete with any of them. It doesn’t post, doesn’t schedule, doesn’t pull platform analytics. It’s upstream of all that: the research and data-gathering layer that feeds the content. That’s an important framing, because if you buy Peji expecting a social media management suite, you’ll be disappointed. If you buy it expecting a research intern who lives in Safari, the value proposition is clearer.
Why TikTok creators should care more than LinkedIn ones
Here’s an opinion, clearly flagged: the creators who’ll get the most out of this are the ones whose content depends on fresh, specific, verifiable facts — deal hunters, travel creators, finance and product-review accounts, anyone doing “I tested X vs Y” content. Those formats live and die on whether the price, spec, or policy you cited is current. A mobile agent that pulls a sourced comparison table in two minutes is a genuine production accelerant for that cohort. A LinkedIn ghostwriter doing thought-leadership posts? Less so — your research is mostly reading, not navigating, and a standard chatbot covers it.
What social teams can borrow from the design
Even if you never install Peji, the product makes three arguments worth stealing for your own workflow.
1. Confirmation gates are a content-safety pattern, not just an AI-safety pattern. The “stops before it sends” rule maps directly onto how you should run any AI-assisted publishing pipeline. If you use AI to draft captions or generate reply suggestions, the human approval step before anything goes live isn’t bureaucracy — it’s the same trust contract. I’ve seen teams wire AI straight into scheduling tools with no gate, and the failure mode is always the same: one hallucinated stat or off-brand reply, screenshotted, viral for the wrong reason.
2. Sourced output should be the default, not a feature. The table-with-links format is what every analytics and research tool should ship. When you build your own reporting — even in a spreadsheet — attach the source URL to every number. It sounds tedious; it saves you the day a client asks “where did this engagement figure come from?” and you have to reconstruct your own work.
3. On-device processing is a positioning choice with real operational consequences. The page says chats, files, and dictation stay on the phone. For creators handling unreleased product info, NDA’d brand assets, or just private DMs, “your research doesn’t leave your device” is a meaningful claim — assuming it holds up under scrutiny, which the launch page doesn’t let me verify. My take: treat on-device as a strong default and still read the privacy policy before you paste anything confidential into any agent.
Where the math breaks
The economics are the open question. “Free on the App Store; Pro adds usage” tells me the free tier is metered, but the page doesn’t say how many tasks, how many tabs, or what Pro costs. Agentic browsing is expensive to run — every task spins up real page loads and model calls — so I’d expect tight free limits and usage-based Pro pricing. If you’re evaluating it, the first thing to check is the actual task allowance against your real research volume. A creator doing ten comparison tasks a week has a very different math than one doing two a month.
Where my judgment says it falls short
Three honest caveats.
First, mobile-only is a real constraint for teams. Social media managers I know live in a desktop browser with 30 tabs, a scheduler, and a spreadsheet open. A phone agent is great for research on the go, but it doesn’t slot into the desktop workflow where most production actually happens. If Peji ships a desktop or web version, that changes the calculus; the page doesn’t say it will.
Second, the “real browser” claim deserves scrutiny. WebKit on iOS is real, but many sites serve mobile-lite experiences, gate content behind logins, or block automated navigation. How Peji handles a paywalled article, a CAPTCHA, or a site that detects non-human browsing is not disclosed. In my experience testing similar agents, the failure rate on hostile sites is where the demo magic evaporates. I’d bet the useful-task success rate is high on clean retail and travel sites and lower on anything with bot detection — but that’s my inference, not a measured result.
Third, it’s a single-purpose tool in a category that’s consolidating fast. ChatGPT, Claude, and Perplexity are all racing toward agentic browsing; Apple and Google are building assistant layers into the OS. A standalone app needs a defensible wedge. Peji’s wedge — mobile, tap-to-confirm, on-device, sourced tables — is coherent, but it’s a feature set that a platform owner could absorb. That’s a risk for the product’s longevity, not necessarily for your workflow today.
Who it’s not for
If you need scheduling, analytics, inbox management, or multi-account publishing, this isn’t your tool — that’s Buffer or Metricool territory. If your research is mostly reading long documents rather than navigating live sites, a standard chatbot is cheaper and simpler. And if you’re a team that needs shared research libraries and audit trails, a single-user phone app won’t fit until there’s collaboration tooling, which the page doesn’t mention.
What I’d watch / test next
This week, if you’re curious: install the free tier and run three calibrated tests. One, ask it to compare pricing across two competitor sites you already know cold — check whether the cited links match the numbers. Two, give it a task that requires a form fill but not a purchase — a newsletter signup or a quote request — and confirm the tap-to-confirm gate actually fires before submission. Three, time a real research task you’d normally do manually (a flight search for a shoot, a PDF summary for a client brief) and compare wall-clock time against your baseline. That gives you a personal success rate instead of a launch-page claim.
What I’m watching from the industry side: whether Peji publishes task limits and Pro pricing, whether a desktop or web client appears, and whether the sourced-table output holds up on messy real-world sites. The broader trend is clear — the browser is becoming the agent’s body, and the creators who learn to supervise an agent rather than do every click themselves will quietly out-produce everyone else. The question is whether a focused mobile app can hold that position before the platforms eat it. My take: the confirmation-gate design is the right instinct, and that instinct is worth copying regardless of what happens to the product.
