Running Ads: From Point-and-Click to Plain Talk

A few days ago, an old friend who runs cross-border ad campaigns invited me out for coffee. He pulled out his laptop and said there was something he wanted to show me.

He said one sentence to the screen: "Show me the ten best-performing ads this week."

The words were barely out of his mouth when the report appeared.

Then he said: "Pause every campaign with a ROAS under 2."

Boom. A dozen-plus campaigns, one second, all paused.

I watched from the side, and after a long moment managed three words: "This is unreal."

Keep in mind: a few years ago, all of this was his midnight routine. Log into the backend, click through layer after layer, change things one by one. Finish Google, move to Meta. Finish Meta, move to TikTok. Clicking until your hand ached, waiting on page loads until you started questioning your life choices.

And now? He says a sentence, and it's done.

So What Is AI Advertising?

Plainly put: you let large models like ChatGPT and Claude operate the ad platforms for you — Google Ads, Meta Ads, TikTok Ads.

To be clear: it isn't the AI giving you ideas. It's the AI actually running things. Creating campaigns, adjusting budgets, pulling reports, adding negative keywords — it just gets done for you.

For instance, you give it commands like this:

"Launch a Performance Max campaign targeting US users, with a daily budget of $100."

"Pause every campaign with a ROAS below 2.0."

"Pull up the ten best-performing ads this week."

"Add every irrelevant search term to the negative keywords."

You speak like a human. It does the work.

This approach has a name: Agentic Advertising.

What's an agent? It runs errands for you. You say one thing, it does one thing. You describe your intent, it understands it, translates it into actions the platform's APIs can execute, and lands them one by one.

How Does It Get into the Backend?

You might ask: AI can't exactly log into the Google Ads backend — so how does it "operate for me"?

Good question. The key is one thing: MCP.

What is MCP? Model Context Protocol. The name sounds intimidating; the idea is dead simple: it's the universal adapter between AI and all kinds of software.

If you want AI to manage your ads, you need something to connect ChatGPT and Claude with the ad APIs of Google, Meta, and TikTok, right?

That adapter is the MCP server.

The market already has products that have turned this into a standard, off-the-shelf component. Adspirer, for one, does exactly this job: install it into ChatGPT or Claude, connect and authorize your ad accounts, and start talking. Google Ads, Meta Ads, TikTok Ads, LinkedIn Ads — all managed inside a single conversation. Without writing a single line of code.

Old way and new way, side by side.

The old way:

1. Log into the Google Ads backend.
2. Jump back and forth across a dozen pages.
3. Click dozens of buttons.
4. Stare at the loading spinner.
5. Switch platforms, do it all again.

The new way:

1. Open ChatGPT or Claude.
2. Say what you want in plain words.
3. The AI executes through the APIs.
4. You check it over and nod.

Same job. Thirty minutes before. Thirty seconds now.

Managing 5 accounts used to mean scrambling. Now you can manage 50 — just a matter of typing a few more sentences.

And it never clicks the wrong button, never tacks an extra zero onto a budget. When it spots an anomaly, it'll call you out even in the middle of the night.

In my head, I've already given this thing a name: a junior media buyer who never gets tired. Watching the data 24/7, fluent in every platform's job all at once — so you never have to learn them one by one.

What If Its Hand Slips?

Hearing all this, you're probably feeling uneasy: what if it wipes out my ad account?

That worry is completely reasonable. Which is exactly why these tools all keep one gate in place: an approval flow.

The AI only proposes; whether anything actually gets executed is your call. It says, "Suggest raising Campaign A's budget to $200." You read it, click confirm, and only then does it take effect.

The steering wheel is still in your hands.

What Depreciates, and What Appreciates?

The technology all works now. But what I keep thinking about lately is something else: the depreciation of skills.

In the past, a great media buyer's moat was knowing where every button was on every platform, and where every trap was buried.

Now the buttons move by themselves.

So what appreciates?

The ability to say things clearly. If you can't spell out "what kind of users I want, what cost I can accept, where my bottom line is," the AI can only guess blind. Same tool: some people use it to mint money, others use it to burn cash like water. What's the difference? Whether you've actually thought your business through.

Back to my friend. As we finished our coffee, he said something that stuck with me: "My most valuable job now is figuring out what's worth putting money behind — not figuring out where the backend buttons are."

This shift was visible to the naked eye starting in 2025. Today, it isn't even a trend anymore. It's just daily life.

The smarter the tools get, the more valuable the people who have truly thought their business through.

Here's hoping an agent soon starts pulling night shifts for your ad account.