I didn't touch Ads Manager. I didn't open my laptop. I read the text as I got my kids ready for school and typed "approved."

That's an OpenClaw agent doing the job. Running my ads account.

I'm giving the whole thing away free. The agent config, three skill, the setup guide. All of it. If you've been waiting for someone to hand you a working AI ad manager, this is it.

Let me show you how it works.

A few months ago I wrote about cutting Meta ads reporting to 45 seconds using an MCP server and plain English questions. That piece got a ton of response. People loved the idea of asking their ad account questions instead of clicking through the Ads Manager nightmare.

But asking questions isn't the gig (just a part).

The job is: pause the loser at 3am before it burns $200. Scale the winner before the window closes. Upload 50 creatives perfectly named, perfectly organized, without a single mistake. Shift budget from dying ad sets to climbing ones.

That's what this agent does.

We didn't ditch MCP. My team is still on it. We deployed it to Railway, added publishing capabilities, got the whole workflow running. MCP is solid. But here's the thing: every MCP call eats your context window. When you want a recurring schedule that runs at 7am every morning (across dozens of ad accounts) whether you're awake or not, that token burn adds up fast. It also requires an active session.

OpenClaw's heartbeats and cron solve that. The agent runs on schedule. No active session needed. Barely touches context. MCP is still in the stack for the interactive stuff. This is the evolution, not a replacement.

Here's what's running in my account right now:

Every morning at 7am, my OpenClaw agent fires the meta-ads skill. That skill wraps social-cli, which is an open-source CLI that talks directly to Meta's Marketing API. No middleware. No protocol overhead. Just commands.

The skill runs 5 questions against live data. Then it acts.

Question 1: Am I on track?

Checks spend pacing against daily budget. If you're 60% through the day with 20% of budget spent, something's wrong. If you're blowing through budget by noon, something else is wrong. The agent flags both and doesn't wait for you to notice.

Question 2: What's running?

Pulls all active campaigns, status, current spend. Gives you a clean picture of what's actually live versus what you thought was live. You'd be surprised how often those two things don't match.

Question 3: How's performance?

7-day metrics by campaign. ROAS, CPA, CPL, CTR. Sorted by what matters. Not buried in Ads Manager columns you have to rearrange every time.

Question 4: Who's winning? Who's losing?

Ad-level data, ranked. Best to worst. The agent shows you the stack. The winners are obvious. So are the ones bleeding money. Then it acts on the obvious ones without waiting to be told.

Question 5: Any fatigue?

This is the one most people miss. CTR trends over time. Frequency (how many times the same person sees the same ad). CPC climbing without obvious cause. These are the early signals of creative fatigue. By the time your CPA spikes, you've already wasted a week. Catching frequency creeping past 3 before the CPA moves saves real money.

7:02am. Telegram message from the agent:

I read that in 90 seconds. Reply "approved" on the scaling rec. Ask about the landing page. Done.

Total time managing ads that day: under 5 minutes.

Here's how the pieces connect.

OpenClaw agent. This is the brain. It runs on schedule via cron and heartbeats. It interprets data, applies decision rules, sends me the brief. It's what turns a pile of ad metrics into "here's what happened and here's what I did about it."

OpenClaw is open source. That's important. You own the whole thing. Not renting SaaS. Not hoping some platform doesn't change their pricing. Your agent, your rules.

Three skills. This is the knowledge layer.

Together they're the difference between raw data and an actual decision. They're what make the agent behave like someone who actually knows paid media.

Clone the repo. The agent picks up all three automatically.

social-cli. This is the connection to Meta. Open source CLI for the Graph API. Handles token refresh, pagination, rate limits. All the stuff that makes Meta's API miserable to work with directly. Honestly go follow and give some love to Vishal Ojha who open sourced this. I was working on something similar, but I saw he open sourced this, I dug under the hood and it’s fantastic (better than what I had).

Those four commands are most of what you need. The agent knows how to use them. The skill tells it when.

Cron and heartbeats. This is what makes it actually run. A cron job fires at 7am. The OpenClaw heartbeat system handles the scheduling without requiring an active session. This is why we're not doing it through MCP alone. The agent wakes up, does its job, sends the report, goes back to sleep. Your context window stays clean.

The full picture:

Cron triggers the chain. The brief lands in your channel of choice.

The skill doesn't just report on data. It does stuff for you!

Pause bleeders. Any ad running more than 3x target CPA for 48+ hours gets auto-paused. No approval needed. The agent notifies you after. You can always resume, but the bleeding stops.

Resume winners. If an ad that was previously paused has a changed context (new audience, refreshed creative) the agent can resume it on instruction.

Shift budget. This one's approval-required for now. The agent recommends a budget shift with the data behind it. You say yes or no in two seconds. It executes.

Fatigue checks. When frequency or declining CTR triggers the threshold, the agent flags the creative and writes the observation to a workspace file. Over time you build a log of what fatigued when and why. Pattern recognition you can actually use.

Custom reports. Ask it anything. "What's my CPL trend over the last 30 days?" "Which creatives are still performing after day 14?" It pulls the data and builds the answer.

Meta just acquired Manus AI. And now they're rolling out AI inside Ads Manager. My team isn't fans (just yet).

Here's the thing about Meta building AI into your ad management: Meta does not optimize for you. They optimize for Meta.

They want you spending more. They want broader audiences because broader audiences mean more inventory sold. They want you running more placements because more placements means more revenue for them. Their AI will always point in that direction.

And look at their track record as a platform steward.

Meta has not proven to be a trustworthy cultural or brand steward. That's not even a hot take. That's history.

So when Meta says "let our AI manage your ads for you," the smart question is: manage them toward what outcome? Yours or theirs?

Your AI agent optimizes for your target CPA. Your cash flow constraints. Your brand guardrails. It has no incentive to burn your budget faster.

The winning move isn't choosing between Meta's AI and your own. Use both. Let Meta handle targeting and delivery (they're genuinely good at that now). Use your own agent to manage the strategy layer on top.

You keep the judgment. They run the auction.

Step 1: Install social-cli (15 minutes)

Test it:

First time you see your ad data in a terminal is a thing. It hits different.

Step 2: Get the OpenClaw agent and skill (free)

I'm giving away the complete setup. The agent config, the meta-ads skill with all 5 daily questions and action commands, and the setup guide. Everything you need to run this in your own account.

grab it here: https://github.com/TheMattBerman/meta-ads-kit

Step 3: Drop the skill into your workspace

The skill goes in your OpenClaw workspace. Point it at your ad account. You're done. The agent picks it up automatically.

Step 4: Set up the morning cron

One cron job. Fires at 7am. Runs the 5 questions. Sends the brief to wherever you want it. Telegram, Slack, email. Your call.

Step 5: Start in review mode

For the first week, the agent recommends. You approve. Get comfortable with what it's seeing and how it's reasoning. Then graduate to auto-pause for the obvious bleeders. Keep the budget shifts in approval mode as long as you want.

You don't have to hand over full autonomy on day one. Start with visibility. Earn trust in both directions.

Not a media buying team. Not creative strategy. Not positioning. Not the calls that actually require judgment.

What it replaces:

The 20 minute sessions clicking through Ads Manager hoping you don't miss something. The end-of-week scramble to figure out why your CPL jumped and whether it was day-of-week or creative fatigue or audience overlap. The extra $1200 that burned overnight because you didn't catch the CPA spike before bed.

That's not strategy brother. That's maintenance. And maintenance is a terrible use of your time.

The 10% that's actually hard, knowing when to go broad versus when to tighten targeting, knowing which creative angle to test next, knowing when the data is noise versus signal, that's what you should be spending your brain on.

Not clicking.

I'm giving away the whole stack. A complete OpenClaw agent with three skills, decision logic, and everything configured. Clone it, authenticate, run it.

You get:

Takes about 30 minutes to get running. Then it runs.

If you'd rather have someone build this into your account and show you what your data is actually saying, that's what we do at my agency. Reply and we'll talk.

Let's build,

Matt

P.S. The Ads Manager UI is still one of the worst pieces of software I've ever regularly used. This doesn't make it better. It makes you not have to use it.

🎁 Get rewarded for sharing! My team grew NEW accounts to over 50 million views in just a few months. We made an AI viral hook generator so you can follow the same hook frameworks that we do.

Invite your friends to join, and when they sign up, you’ll unlock our AI Viral Hook Generator—the ultimate tool for creating scroll-stopping content.

{{rp_personalized_text}}

Copy and paste this link: {{rp_refer_url}}