Programmatic Was the Beginning. So What Happens When Advertising Becomes Agentic?

I've been thinking about the way advertising automation is evolving.

And the more I think about it, the more I wonder whether we're looking at a completely new revolution or simply the next layer of something that started a long time ago.

Because, in many ways, 'programmatic' was already the beginning of AI-like automation.

We just didn't call it that.

Programmatic changed the way we bought and sold advertising

Instead of people manually deciding which impression to buy, at what price, and for which audience, technology started doing it for us.

Data became part of the decision.

Algorithms started making decisions.

Transactions happened in milliseconds.

Budgets could be optimised automatically.

The human was still there, of course.

But we were no longer doing every single thing manually.

We were telling the system what we wanted and letting the technology execute.

Sound familiar?

It should.

Because now we're talking about agentic AI.

And I think this is where things get really interesting.

We are not starting automation from zero

When I hear people talking about agentic AI in advertising, sometimes it sounds like we're suddenly moving from a completely manual world into an automated one.

But that's not really what happened.

We've already spent years automating parts of the advertising ecosystem.

Programmatic automated transactions.

Bidding systems automated decisions.

Optimisation algorithms automated budget allocation.

Audience technologies automated targeting.

Ad servers automated delivery.

So perhaps the next step isn't:

human → automation

It's:

automation → autonomous automation.

And that's a very different thing.

The interface is changing

Today, you might still need someone to go into a platform and set up a campaign.

Choose the audience.

Choose the inventory.

Set the budget.

Set the bid strategy.

Set the optimisation goal.

Check the results.

Make changes.

Repeat.

With agentic AI, the interface could become much simpler.

Instead of setting everything up, perhaps I just say:

"I want to increase incremental revenue from this audience. Use this budget, prioritise profitable outcomes and move spend towards what's working."

And the agent figures out the rest.

It could build the campaign.

Choose the inventory.

Adjust the bids.

Move the budget.

Test different approaches.

Analyse the results.

And keep going.

That sounds like a huge change.

But here's the part I'm interested in.

The human hasn't actually disappeared.

The human has just moved.

From operator to decision-maker

Maybe the biggest change isn't that AI takes humans out of advertising.

Maybe it takes humans out of the execution layer.

We move from:

"How do I set this campaign up?"

to:

"What do I want this system to achieve?"

That's quite a big shift.

But it also creates a much bigger responsibility.

Because telling a machine what to do is not necessarily the same as knowing what you want.

If I tell an agent:

"Optimise performance."

What does that actually mean?

More revenue?

More profit?

Higher ROAS?

More customers?

Better customer lifetime value?

More attention?

More market share?

The AI can execute the instruction incredibly quickly.

But it still needs someone to define what success actually looks like.

And that's where I think we sometimes underestimate what is happening.

The prompt is not the strategy

A natural-language instruction can make something look incredibly simple.

But underneath that simple prompt could be hundreds or thousands of decisions.

Which data should be used?

Which data should be ignored?

Which audience should be prioritised?

What level of risk is acceptable?

When should the system experiment?

When should it stop?

How much budget can it move?

Which publishers should it include?

Which should it exclude?

When should it ask a human?

Those aren't just technical questions.

They're business decisions.

And if the AI is making them, someone has delegated that authority.

Which brings me to the question I think we're going to have to talk about much more.

Who is responsible when it goes wrong?

Let's say an AI agent is told to optimise towards profitable growth.

It decides that a particular publisher isn't delivering enough value.

It moves £500,000 of spending elsewhere.

The publisher loses significant revenue.

The advertiser doesn't get the expected incremental growth.

The AI says:

"Based on the objective and available data, this was the optimal decision."

But the outcome was wrong.

Who owns that decision?

The person who wrote the prompt?

The company that deployed the agent?

The agency?

The DSP?

The AI provider?

The data provider?

The person who approved the system?

This is where I think the conversation around agentic advertising gets much more interesting.

We're asking:

"How autonomous can we make AI?"

Maybe we should also be asking:

"How much responsibility are we prepared to delegate?"

Because autonomy without accountability is a very uncomfortable combination.

Maybe the human moves up the stack

I don't think the answer is to keep humans involved in every single decision.

If we're still manually approving every bid, every audience and every budget adjustment, then what's the point of having an agent?

Instead, perhaps the human role becomes something different.

We define the objective.

We define the boundaries.

We decide what the system can and cannot do.

We decide when it needs approval.

We decide what evidence it needs.

And importantly, we decide who is accountable.

The machine can execute.

But someone still needs to own the outcome.

That's where I think the architecture becomes more important than the agent itself.

And this is where I see the next opportunity for media

I've spent most of my career somewhere between the commercial and technical sides of advertising. I've seen programmatic infrastructure evolve.

I've worked with publishers, SSPs, DSPs, OpenRTB, first-party data, yield optimisation, integrations and monetisation.

I've also seen that having the technology doesn't automatically mean you have a good commercial model.

Sometimes the technology works beautifully, but the business model doesn't.

Sometimes the data is there, but nobody knows how to turn it into revenue.

Sometimes the inventory exists, but the infrastructure isn't built to monetise it properly.

And sometimes we automate a process that probably shouldn't have been designed that way in the first place.

That last one worries me the most.

AI can make a bad system much faster.

So I don't think the next challenge is simply implementing AI.

It's understanding what we are actually building around it.

This is where Quintworx comes in

This is increasingly how I think about Quintworx.

Not as another company helping businesses "use AI".

And not simply as another programmatic consultancy.

The interesting work is further upstream.

Looking at the whole system.

The commercial objective.

The media inventory.

The data.

The technology.

The operating model.

The people.

The automation.

And ultimately:

the outcome.

Because if we're entering a world where AI agents can execute media decisions with very little human interaction, businesses need to understand how all those pieces connect.

What should be automated?

What shouldn't be?

What should an agent be allowed to decide?

Where should human judgement remain?

How do we measure the actual outcome?

And who is accountable when the system makes the wrong call?

That's not just an AI question.

It's a media architecture question.

So, have we really changed the whole picture?

I'm still not sure.

Maybe we're not replacing human advertising with AI advertising.

Maybe we're simply moving through another stage of the same evolution.

First, we automated the transaction.

Then we automated the optimisation.

Now we're starting to automate the operator.

And perhaps the next stage is where the human moves further up the stack — from doing to directing.

That could be incredibly powerful.

But it also means we need to get much better at defining objectives, setting boundaries and taking responsibility for outcomes.

Because the future isn't necessarily:

Human vs AI.

It might be:

Human intention. AI execution. Real-world outcomes.

And somewhere between those three sits the architecture that makes the whole thing work.

That's the bit I'm interested in.

Because the future of media isn't just about making machines smarter.

It's about making the system around them smarter too.

— Quintworx
Revenue Architecture & Media Transformation