The case against AI agents for programmatic ad buying

Yes and no. That may seem like a mixed answer, but it was clearly made by agency executives throughout last week’s Digiday Programmatic Marketing Summit in New Orleans, La. 

“There’s a role for AI in general and a lot of different avenues, but I think when it comes to programmatic activations, we’re not trusting that to large language model-based agents,” said Christopher Francia, director of product development and client performance at Attention Arc, in a live recording of the Digiday Podcast during DPMS.

Hallucination is the primary reason for hesitation. If an AI agent errs by a decimal point, that could not only be financially catastrophic but a fireable offense (and it wouldn’t be the AI agent getting the axe).

But there’s also the question of whether AI agents would be able to work quickly enough to function in real-time bidding auction. Initiatives like IAB Tech Lab’s Agentic RTB Framework seem designed to help speed things up, but they require widespread adoption.

None of that is to say that AI agents don’t have any place in programmatic advertising workflows. But for the moment they’re best relegated to lower-stakes tasks, like campaign brainstorming and insight summaries.

“When it comes to your agentic stuff, it’s definitely more on the insight, analysis and ideation side,” Francia said.

Here are a few highlights from the conversation, which have been edited for length and clarity.

The main case against AI agents

We don’t tell ChatGPT, “Go and set up this campaign and activate this campaign and select the best targeting options” for a lot of logical reasons. Mostly, it’s going to hallucinate. If there’s an error that gets triggered because some condition wasn’t known ahead of time — it’s just too many variables for an AI to do. AI is really good at very specific, narrowly tailored tasks. The more complicated a task becomes, the harder it is for it to figure out what it’s supposed to be doing.

The creative case against AI agents

If you’re talking about AI modifying creatives in real time in the bidstream, that’s going to be a huge “no” for a lot of brands. Because hallucination. It just takes one mistake for that brand’s 100-year legacy to have a huge crisis. We had a company pitch us this, and I said very plainly, “Sounds like a really novel idea. No brand on earth that has a reputation is going to risk that because they have no control.”

The speed case against AI agents

When [people are] talking about AI in the bidstream, they’re mostly talking about AI talking to demand-side platforms. But they can’t talk to the bidstream. And the reason there is physics. The bidstream is a very small window of time to communicate with it. I think a DSP has maybe up to 100 milliseconds. Your fastest LLM can’t even get close to that.

The case for AI agents in programmatic workflows

AI is good for insights and summarization and helping you ideate. We found a lot of success with AI [with prompts such as] “What are some potential audiences that might be relevant to this [audience behavior insight]?” It might not get it all right, but it helps us ideate. Other ways we utilize it is menial tasks like “Look, I need to copy this campaign template 10 times, and I need to make these changes in it each time. Can you just do that for me?”