You open your phone on a Monday morning and a Meta ad campaign already ran overnight. You didn't build it. Nobody on your team logged into Ads Manager.
An AI agent picked the creative, set the budget and launched it while you slept. It's already spending money by the time you've had coffee.
That's not a thought experiment anymore. On November 3, Perplexity shipped e-commerce workflows for its Computer agent. The pitch is blunt. Perplexity says Computer can analyze a store's Shopify sales, build the landing page, run the Meta ad and send the Klaviyo email, all from one request.
What Perplexity's Computer actually does now
Perplexity Computer already handled coding, scheduling and research for Pro and Max subscribers. The new piece connects three accounts. Shopify supplies store data and landing pages, Meta runs the ad campaigns, and Klaviyo sends the email. A brand can save its design guidelines and voice as a reusable skill, so every page, ad and email pulls from the same rules instead of drifting.
Perplexity's own example shows how specific the starting brief can get. Perplexity wrote the brief itself, and it says: "Promote our new summer collection with a 15% discount, a two-week Meta ad campaign capped at $1,000, and a Klaviyo email campaign for existing customers. Show us all materials for review before launch." A team decides, up front, which of those three actions (publishing, sending and spending) need a human to sign off before Computer acts.
This is a step past what we covered when Perplexity Computer for Builders launched back in August. That version wrote code and watched production metrics for solo developers. This one runs jobs most online stores already pay an agency, a freelancer or a tired founder to handle every week.
The $30,000 test nobody's audited but everyone's quoting
Perplexity backs the launch with two customer numbers, and the bigger one is moving fast online. For Get Maine Lobster, Computer managed a $30,000 Meta ad budget. Each day it scanned competitor ads and checked how the account and the Shopify store were doing.
It picked that day's creative, cut new versions of the best videos, and adjusted the budget. Against an agency-run campaign over the same spend, Perplexity reported 3.5 times the revenue and a 63% jump in return on ad spend.
That comparison is worth reading carefully. It's Perplexity's own, published on its e-commerce solutions page to sell the product, not a figure an outside auditor checked. Unite.AI reported the same figures straight from that page, and neither write-up says what the agency campaign targeted, how it was paced, or whether the two ran in the same season. The number is still real. Read it as a vendor's best case, not a lab result.
The second number sits on the Klaviyo side. True Classic's lifecycle marketing lead, Sivabalan Rajasekaran, told Perplexity that Computer rebuilt 19 customer-experience flows in a few hours. He said the same work would normally take weeks and cost thousands of dollars in agency fees.
200+ SaaS teams already track their AI citations.
They know exactly when ChatGPT mentions their brand, and when it stops. Do you?
Why your agency should read the daily loop twice
Here's the part that should worry freelance ad buyers and small agencies more than it worries marketers in general. That daily loop for Get Maine Lobster, scanning ads, checking performance, picking creative, adjusting budget, reads like a job description. It's the job a large share of people now bill hourly or monthly to do on someone else's ad account.
Agentic commerce is the broader trend this sits inside. These are AI agents that act inside real accounts, with real money, rather than only answering questions about a brand.
One marketer posted online, asking whether anyone would hand an AI their ad budget without a person watching nearby. That's a fair question. For most teams the honest answer is a little, carefully, with every approval gate turned on, which happens to be how Perplexity built it.
There's a second layer most people are missing. Computer's loop starts by scanning competitor ads and checking performance. That means what's publicly visible about a brand, and about its rivals, now feeds a budget decision directly, well beyond shaping a chatbot's answer. AI visibility tracking used to mean checking whether ChatGPT mentioned a company correctly. Now it can also decide whose ad budget goes up.
What the approval gates cover, and what they don't
The three gates Perplexity offers are specific. A team can require review before publishing a page, before sending an email, and before spending on ads. Turn all three on, and Computer builds everything, then waits for a human before any of it goes live or gets paid for. Turn one off, and that action happens on its own from then on.
What the gates don't cover is everything upstream of those three moments. Competitor research, copywriting, image generation and the first draft of a plan all happen on their own either way.
By the time a human reviews anything, Computer has read a brand's competitors, decided what to say about a 15%-off summer sale, and written the email. The review step is real. It just reviews finished work, with nobody at the table while that work gets made.
A few things are worth doing before connecting Shopify, Meta and Klaviyo to any agent, Perplexity's or otherwise:
- Read the brand skill Computer builds from before trusting it. Thin design notes just mean the agent fills the gaps with its own judgment.
- Keep every gate on for the first few campaigns, then loosen the ones that keep producing clean drafts.
- Set the budget cap inside the brief itself, the way Perplexity's own example does, rather than trusting a default.
- Treat the first week of output as an audit, not a launch, and read every ad and every email before it reaches a customer.
How often does ChatGPT mention your brand?
Most founders have no idea. The answer might surprise you.
Approving the first campaign is the easy part. Knowing, six weeks in, whether the agent's read on a brand and its rivals is still accurate is the part most teams aren't set up to check. Nobody re-reviews draft four hundred the way they reviewed draft one.
A team can watch every draft these agents produce by hand, every week, for as long as the accounts stay connected. Or RankControl can track the layer below it.
Across seven AI engines, we check every week whether those engines still describe a brand and its rivals the way they did when the accounts went live. That frees the team to focus on the campaigns themselves.
BFCM lands three weeks after this launch, right when a store owner has the least spare time to babysit a new tool. Every gate Computer offers has to be set by the team connecting the accounts, so the only thing standing between an agent and a store's ad budget is whichever boxes get checked the day it's connected.
15 hours a week manually. Or 15 minutes with RankControl.
Track citations, monitor competitors, and fix content gaps across every AI search engine. Automatically.
