AI was the silent engine of programmatic advertising for years. Not sexy, but effective. Smarter bidding, better targeting, faster optimization. But 2026 feels different. AI won't just come to optimize, AI will be at the table. Not as ‘another algorithm’ but as an agent. A system that can email itself, negotiate, set appointments, set up test deals, adjust and feed back.
Welcome to Agentic Advertising.
Why this is playing out right now
Everyone who works in programmatic recognizes the pattern. The market shouts something new, everyone shares a hot take and before you know it, you're in a meeting where someone asks: “Should we do something with this?”
Agentic, AdCP, autonomous deals, it sounds like yet another buzzword bingo. At the same time, you can also see why this is gaining traction. The pressure is on: faster switching, less manual work, more control, better performance. Sounds like something we like in busy times. And meanwhile, the playing field is getting more complex. Brand safety, privacy, supply path, curated deals, reporting, you name it.
The question is no longer whether AI helps. The question is how much responsibility will you give AI.
But what is Agentic Advertising?
Agentic Advertising is all about autonomous AI agents that perform actions independently. They can communicate, negotiate and perform transactions without a human having to press enter all the time.
In programmatic, this means very specifically:
- On the advertiser side runs an agent that understands your brief.
- On the publisher side runs an agent who monitors inventory, floors, deal terms and availability.
- Those agents ‘talk’ with each other and convert deals into actions.
So not just: “optimize my campaign”. But also: “arrange an appropriate deal for this briefing, test it, scale up if it works and explain why.”That sounds futuristic, but the first pilots already exist. Consider a process where an agent, based on a briefing via email or API:
- propose a deal setup,
- checks terms,
- a test flight starts,
- analyzes results,
- and then makes adjustments.
The promise is clear: speed, fewer errors and less manual hassle.
And then AdCP comes around the corner
AdCP stands for Advertising Communication Protocol. The idea is simple: if agents are to work together, they need a “language”. A standard way to communicate about things like inventory, pricing, targeting, brand safety requirements, delivery agreements, and reporting.
I understand the comparison with HTTP, but I want to put nuance on that right away: HTTP is mature and universal. AdCP is far from that.
See AdCP now mainly as an ambition, an attempt to make this kind of agentic process scalable, even between different parties and systems.
What is often mentioned as advantages:
- Transparency: appointments and changes are easier to log and retrieve.
- Interoperability: less reliance on één platform or één party.
- Efficiencies: faster from briefing to test and live.
But to be fair, this is still in the “proof of concept” phase. Interesting, promising, but not yet a standard that everyone will be running on tomorrow.
The human touch, right here
AI is good at what is measurable. KPI’s, pacing, CTR, CPM, viewability, conversion rates. Top. Only advertising is more than measurable. Context, brand identity, reputation, creative feel, and the question: “does this fit our brand, in this place, today?”
An agent may well decide that a particular environment “converts” just fine. Until your client calls because their brand is next to a sensitive news story. Or because the campaign ran perfectly technically, but totally missed the mark strategically.
This is why I don't believe in “AI takes over”. I believe in “AI takes over work”. A lot of work even. But the direction remains human.
At Media House, for example, we are experimenting with agents who can make proposals for deals based on a briefing. Super useful, especially to get to options faster and have less manual and repatriative work. That makes work more fun and less error-prone, so automation is to be loved. Only we keep the control with the human, the human touch in AI. The planner, the strategist, the marketer you are that human touch. Technology supports, but does not control.
What you can do right now, without going crazy
.
You don't have to wait until this is fully mature. This is precisely the time to test, play and thereby develop yourself with clear frameworks.
This is how I would go about it:
- Choose ééone simple use case
For example: generate deal proposals based on briefings, or automate floors and inventory checks. Don't immediately “let the agent do everything”.
- Work with clear guardrails
What is the agent allowed to do, and what isn't? Consider maximum CPM, excluded environments, minimum fire safety requirements, and escalation rules.
- Test small and short
Eén advertiser, één publisher group, één format, two weeks. Then evaluate. If it doesn't work, you learned quickly without damage.
- Make sure audit trail
Record why decisions were made. Not only for compliance, but also because otherwise you will later have no idea why something suddenly went differently than expected.
- Measure not only performance
Also look at quality: context, supply path, stability of delivery, and how much manual labor you really save.
Where this is likely to go
2026 is going to be a year of transition. You're going to see headlines like “AI buys media better than humans”. And yes, in a narrow definition, that can sometimes be true. If “better” means: optimizing faster on één KPI.
But practice is more recalcitrant. That's why I expect a hybrid model:
- Agents that autonomously negotiate, set up, test and adjust.
- People who set strategy, creation, ethics and brand frameworks.
The marketer of tomorrow works not less, but differently. More as a director of smart agents.
Embrace it, but stay on the wheel
Am I afraid that agentic advertising is taking over my job? On the contrary. I'm especially looking forward to what it will bring if we deploy it well.Less hassle, faster to the core, and more room for the work that really adds value. Strategy, creativity, relationships, and making choices beyond a dashboard. New business can feel like uncomfortable and hard work, but as a good friend once said to me as I cycled exhausted, cursing and plodding up a mountain on a tropical island: embrace it. Do the same here. Test, learn, adjust. Agents don't take the wheel, they help us navigate faster and smarter. And that is exactly what Mediaminds stands for: change with knowledge, vision and a touch of guts.
