Perplexity Computer replaced $225K/yr in marketing tools in a single weekend. We built an AI marketing agent that scans hourly, manages budgets, detects fatigue, and coordinates several campaigns end to end. In one test run, it made 224 micro-optimizations to our ad stack.

@RoryCrave Please stop shipping, I have a family to feed

A great demo that will not work in actual production. Media buying agents have existed in demo form for the last 2 years, the issue has not been connecting to the Meta API and letting LLMs run random strategies. This is literally just a more complicated version of the ads manager itself. As someone that's worked with 100s of clients to help them get an ROI from Meta ads....... the business manager itself is already complex as it is, 90% of biz owners can NOT navigate it & still can not install a pixel. Creating another fancy AI ads manager + dashboard solves absolutely nothing. This is just another tool people will need to learn, debug/diagnose and optimise. Not going to happen. Another fancy tool that absolutely no one is going to use or get ROI from.

Leave something for us to do

idk who made this video, but the first screen makes it feel like a trading screen, and then you read that it replaced a marketing agency. def a disconnect and not a good marketing job!

"224 micro-optimizations" , we've ran millions in ad spend profitable, no one ever runs '100s of micro optimisations', that's NOT how ads work at all , lol

Seeing a lot of companies being banned without notice by Meta for automating Meta ads. What is your take on this, @AskPerplexity?

This stripe billing issue has existed on your checkout a week+. Can you all fix it?

Computer, make no mistakes.

Every corporate employee right now:-

Another fancy tool to burn through your ad spend money. lol for people who think this will work, go try it out and then come back to this post.

I'm calling CAP with this crypto ass looking dashboard mascarading as "Meta Ads".... Boyyyy if you don't get tf out of there with that b.s.....

@grok explain to me what perplexity did here and whether it was effective

224 micro-optimizations in one test run. We did 11,000 last Tuesday. Still spent $312K that month. We're not entirely sure on what. The attribution model attributes it to "miscellaneous agent overhead." Nobody has investigated.

Everyone except Manus will probably be banned from FB ads

I am convinced a lot of people building all these fancy AI solutions for advertising actually have 0 knowledge of advertising and what is most important. So it did 224 things...cool! What did it actually do? How did it improve performance? Was it even necessary to make the optimization in the first place? Who trained is on when to make an optimization, what the optimization should be and how to determine whether it was the right decision? Until one of these posts also includes the actual data on performance improvements (I am talking dollars) they should be ignored. It is like the people back in the day who would share screenshots of their ads managers to prove how good their strategy was and only show one ad or not show the ROAS column.

That's very cool. Any insights on conversion after those micro-optimizations?

What Perplexity is building doesn’t hold as much long-term value as it might seem. The real holy grail has always been base model intelligence, and Perplexity doesn’t own that. They’re building software layers on top of borrowed AI models, and in an era where building software is becoming trivial, that’s not a defensible position. Their features will be dead on arrival the moment the underlying model providers ship the same capabilities natively.

The real win isn't the $225K saved—it's the 224 micro-optimizations in one test run. That's the feedback loop humans can't match anymore.

send me the prompt/trace!

The micro-optimizations part hits hard. Built similar automation for my Twitter tool and watching it make 50+ adjustments overnight that would've taken me hours was wild. The real value isn't replacing humans - it's freeing us up for the strategic decisions only we can make.

The 224 micro-optimizations number is the one to focus on. Not because its impressive but because no human was doing those. They weren't replaced. They just weren't happening. AI in marketing isnt replacing what humans did. Its doing the layer of work that was always too small to prioritize

i spent all day yesterday trying to fix a single ad budget error so seeing a bot do 224 optimizations in one go is pretty wild.

@grok which tools can be replaced?

wow, everything is computer

hourly scanning, budget management, fatigue detection, and end-to-end campaign coordination replacing $225K worth of tools is the ROI calculation that makes the $200/month Computer subscription the easiest budget approval of 2026.

Humans tweak those once a week. Agents can adjust them every hour. The real shift isn’t “AI does marketing.” It’s that optimization becomes continuous instead of periodic.

It’s great but I am smarter now than a month ago where I would gasp at these launches and my timeline will be filled with “under cover” influencers talking great things about these launches and the features and the benefits. However now I can smell from distance whether it is a genuine appreciation post or a paid or mutual benefit campaign. It becomes so obvious that suddenly 40-50 people taking great things about new launch. No diverse views / opinions. This whole thing works in fooling people once. Not everytime.

cool post but this is like a junior chef adding salt grain by grain to a dish and thinking that's clever

The amount of additional credits you would need purchase to replace a “$225K/yr in marketing tools” would exceed $100K/user. Perplexity Computer is the Detroiters poop car of AI platforms

Can you guys get Google and Meta to allow this? Seems like when it get stuck in a loop you can get banned

omg

replacing a $225k stack in a single weekend is the ultimate dev flex. moving from hiring a growth lead to spinning up a docker container that doesn't sleep or ask for equity is the new meta

The $225K number is impressive but it's not even the real story. The real story is that 224 micro-optimizations happened hourly. No human marketing team does that. They check dashboards once a day, maybe twice if the CEO is breathing down their neck. AI agents don't just replace tools. They replace the entire feedback loop speed that humans physically can't match.

$225K claim is marketing. The real test is whether it makes strategic decisions or just shuffles budgets.

A great demo that will not work in actual production media buyin agents have existed in demo form for the last 2 years the issue has not been connectin to the meta api and letting llms run random strategies this is literally just a more complicated version of the ads