I handed the machine the keys to a live ad account and asked it the only question that matters. What is working, what is bleeding, what do I kill.
A Dubai venture brand I run growth for. Real spend, a stack of ads, a client waiting on results. So I gave Claude access to the whole thing and let it read every ad, every number, every angle. In minutes it came back with a clean report: these are pulling, these are dead weight, here is what to change.
Fast. Confident. Impressive. And not enough.
The read was half baked
The analysis saw the patterns in the data. It could not see the things that never make it into the data. It told me which ad was weak. It could not tell me why a Dubai buyer was actually scrolling past, or which creative angle that buyer would trust, or what to put in the dead ad's place so the whole account started breathing again.
That part is not analysis. That is seven years of running ads and watching real people behave in ways no dashboard predicts.
So I took what the machine found, ran it through everything I have learned, and built the actual fix. Not the one in the report. The one that works. The client is happy now. The problem is solved. The ads are finally doing their job.
The same week, Mastercard launched Agent Pay for Machines. AI agents can now transact on their own, at machine speed, down to fractions of a cent. InMobi and Scope3 shipped an agent that buys media with no human in the loop. Two-thirds of ad buyers say they are now chasing agentic AI for buying and execution. The whole industry is in a hurry to hand the credit card to the robot.
Machines can spend your money now. The real question is whether you want them to.
Sit with that. If I had let this one act on its own, it would have scaled a fast, confident, half-right decision with my client's money. Instead it did the heavy lifting and I did the judgment. That order is the whole job.
I saw the dream version this week too. A reel crossed six hundred thousand views where a creator hands the mic to his own AI agent. The agent narrates the business numbers for fifty seconds, then drops the kicker: "the static carousel was weakest, so I paused it." Everyone shared it as proof the machines have arrived. They have. But that line only works because somebody, once, taught it what weak means. The judgment came first. The autonomy came after.
Most teams are automating the wrong half
Gartner said the quiet thing out loud last month. AI layoffs may free up budget, but they are not delivering returns. Read that twice. Companies are cutting people and keeping the busywork, then wondering why the spreadsheet did not get smarter.
That is backwards. You do not agentize the judgment. You agentize the admin.
My own stack is built on exactly this split. The boring runs on its own:
- End-of-day work logs write themselves across every brand I touch.
- My second brain files the chaos so I never lose a thought.
- A finder watches for reels going vertical so I never start from a blank page.
- This post you are reading got pitched to me by a machine before a human ever typed a word.
None of that needed a person. All of it used to eat my evenings. So I deleted the evenings, not the people.
But the call on a client's budget stays mine. The sixth cut of a launch film stays mine. The decision to kill an ad and choose what replaces it stays mine. Those are not admin. That is taste, and taste is the one thing I am actually being paid for.
What this means if you run a brand
Stop asking "what can I automate." Start asking what is the boring half and what is the judgment half. Then automate one and protect the other with your life.
Let the machine analyze all day. Let it read your whole account and tell you what is working and what is not. That is a real gift, take it. Then put your own experience on top before anyone spends a thing. The analysis is the first draft. You are still the editor.
Everyone is about to learn the hard way that "end to end" never meant "unsupervised." The brands that win the next year will not be the ones who automated the most. They will be the ones who automated the right half and kept their sharpest human on the other one.
The machine read my whole ad account in minutes and it genuinely helped. It just could not finish the job. I did, with seven years of scars. My client is getting results because a human read what the machine missed. Machines can spend your money for you now. Whether you let them is still the most expensive decision you will make.
I write more of these in Hyperthoughts: culture, AI, craft, and the future of the work. From Dubai.
Abhileen Singh Saluja · That Hyperactive Sardar. Find me: Website · Instagram @abhileen.ai · LinkedIn · Reach out