Everyone Loves AI Until It Touches the Budget
Everyone’s fine letting AI plan campaigns. No one’s ready to let it actually spend the budget (yet).
I read the 𝐃𝐢𝐠𝐢𝐃𝐚𝐲 piece about AI buying agents and suddenly the industry behaviour made complete sense.
We’re not scared of AI. We’re scared of AI with a credit card.
Planning agents are having a lovely time right now.
They read briefs. They suggest audiences. They build keyword lists. They organise PMPs. They tidy up workflows everyone secretly hates.
Omnicom’s testing them.
Kinesso’s testing them.
SBS is testing them.
Everyone’s nodding along saying “this is great actually”.
Equativ’s Maestro is a perfect example.
It does the boring bit.
Reads the brief. Suggests deals. Packages inventory. Then it just stops.
Like a well trained intern. No spend. No execution. Just vibes and efficiency.
Forty percent planning time saved apparently... which checks out because half of planning is copy pasting logic you already know.
It’s the safe zone tbh... if it messes up, worst case you look stupid in a deck. No one gets sacked or shouted at.
Buying agents though… that’s where the mood changes.
Because now we’re talking about software actually spending client money.
Not recommending. Not suggesting. Executing.
Butler Till testing Scope3’s buying agent is where this gets spicy.
That agent talks directly to a publisher side agent via AdCP.
No DSP interface.
No comfort blanket UI.
No “just pause it if it looks weird”.
It’s proper machine to machine buying.
The promise is seductive. Cut out middlemen. Reduce fees. Strip waste. 40% cheaper execution...
And every media lead, including me, reading that immediately thinks “Cool… who’s responsible when it goes wrong?”
That’s the real blocker. Not the tech. The accountability.
Everyone’s happy trusting ChatGPT to help plan Christmas dinner. No one wants it actually ordering the food, choosing the shop, charging the wrong card, and delivering twenty turkeys because it misunderstood “feeds six”.
Hallucinations are funny until they show up on an invoice.
Then there’s governance.
If an AI overspends, misallocates budget, buys sh*t inventory, or just does something technically correct but strategically stupid… who owns that?
The agency? The vendor? The model? Good luck explaining that in a post campaign call.
And then there’s the bit no one wants to say out loud... but I will... obviously!
Buying is the last bit of the job that still feels human.
Planning has already been abstracted. Strategy decks, audience logic, channel splits… all ripe for automation.
Execution is where traders still justify their existence.
The second an LLM is trusted to activate spend, a lot of job descriptions quietly disappear.
So when people say “I don’t trust LLMs to buy media yet”, what they really mean is “I don’t trust anyone to explain this when it breaks”.
The protocols aren’t ready either.
AdCP is new.
Scope3 are keeping pilots tiny on purpose. Because one bad headline here doesn’t just hurt a test… it sets the whole category back years.
So here’s how I see it playing out... if you’re interested?
Agents that set the table become normal. Brief analysis. Deal creation. PMP organisation. Workflow automation. That stuff becomes boringly standard.
Agents that eat the meal… actually spending the money… stay human gated for a while.
Approval layers. Hard caps. Kill switches. Finance teams hovering like nervous parents at a teenage party.
This isn’t fear.
It’s just people who’ve been burned before remembering what happens when “set and forget” ... well... forgets.
Because anyone who’s ever left a campaign uncapped over a weekend knows EXACTLY why the Buy button is still sacred.
Oh btw... if you actually want to understand how this machine works under the hood… I literally wrote the book on it… WTF IS PROGRAMMATIC?
This took me ages to research and write. All I ask is that you hit subscribe above^ if you found value.
k, thanks, bye