Most discussions about agentic advertising start with execution. I think the useful bit happens earlier.
Give an agent a brief, connect it to a DSP, add a few guardrails and let it build a campaign. It makes for a good demonstration. I’m less convinced it solves the most valuable problem.
Programmatic execution is already highly automated. Nobody is deciding whether to bid $3.72 for impression number 14,382,291. Machines have been doing that for years, so putting another machine above the existing machine may remove some workflow but it doesn’t fundamentally change advertising.
Before anyone presses buy
An advertiser has to work out who it wants to reach and what those people are worth. Then it needs to know where they can be found, how much suitable inventory exists, what a realistic outcome looks like and what should be sacrificed when the budget inevitably fails to cover every ambition in the brief.
Answering that across the open internet remains remarkably difficult. The information is spread between agencies, publishers, buying platforms, measurement companies, identity providers and dozens of specialists. Every system knows something. Very few understand the whole problem.
This is where agents become interesting. They can interrogate a lot of information and reason across competing constraints in a way that makes a complicated market easier to understand. Being a slightly faster bidder feels like a fairly modest ambition by comparison.
Imagine the brief is to reach environmentally conscious parents aged 30–45 across video and display. It asks for quality journalism, premium entertainment, incremental reach beyond social and sensible frequency. The budget is $4 million, because fictional briefs shouldn’t have to deal with procurement.
An agent could test whether the audience and inventory actually exist, where they overlap and what happens when frequency drops from five impressions to three. It could compare publishers, apply the client’s quality rules, challenge the expected price and show where the plan will struggle. By the time anything reaches an activation platform, most of the commercially important decisions have already been made.
The $4 million question
Adtech has a habit of treating any automated workflow as agentic. A natural-language interface can be useful. Automated campaign set-up can save time and nobody will mourn the manual creation of another deal ID. These are worthwhile improvements to an existing process.
Saving a trader two hours creates operational value. Finding out that the media plan won’t deliver before $4 million is committed creates something rather more valuable.
Advertising should care much more about the second. It tackles the quality of the investment, not just the amount of labour needed to enter it into a platform.
Another place to log in
For 20 years, Adtech businesses have built interfaces around their own products. DSPs built buying tools, SSPs built deal tools, measurement companies built reporting products and data businesses built audience builders. Every company naturally assumed the customer wanted another place to log in.
Their agent may not want to use the software at all. It wants access to what the product knows and what it can do. The product question changes: can another intelligence system understand the data, the constraints and the capabilities well enough to use them?
APIs and common semantics suddenly matter more than another dashboard redesign. That’s a much larger architectural change than putting a chatbot inside an existing product and giving the release a suitably futuristic name.
Decide, then execute
Agents will execute parts of advertising. They’ll create campaigns, negotiate packages and move budgets. The work will still sit inside financial controls, brand rules, measurement requirements and existing optimisation systems. That’s probably a good thing when the system has access to somebody else’s money.
Real-time bidding operates in hundreds of milliseconds. A language model has no reason to make every auction decision itself. The agent can decide the strategy and pass an instruction into infrastructure designed to process billions of transactions efficiently.
The competitive race moves upstream to the information shaping that instruction: audience evidence, supply intelligence, delivery forecasts and a credible comparison of the options. Whoever influences the recommendation sits very close to the advertiser’s budget.
The first genuinely valuable application of agentic advertising may never place a bid. If it can tell an advertiser what to buy and why, with more confidence than the current process, it will have changed far more than another automated campaign builder.