AI Agents are Coming for Adtech’s Tedious Work, Not the Upfronts: Magnite’s Paige Bilins
AMENIA, N.Y. – Advertising technology has spent decades finding increasingly sophisticated ways to automate buying and selling media. Now artificial intelligence is promising to automate some of the work involved in operating all that automation.
Paige Bilins, senior vice president of product management at Magnite, sees potentially significant changes ahead as AI agents become capable of making decisions and acting on them. But she cautions that much of what the industry currently calls “agentic AI” is still closer to good old-fashioned automation with a shinier name.
“We’re still in very early innings around AI,” Bilins said in an interview with Beet.TV contributor David Kaplan at the Beet Retreat Berkshires.
Agentic AI isn’t just faster automation
Bilins draws an important distinction between AI automation and genuinely agentic systems.
Automation takes an existing task and performs it faster or more efficiently. An AI agent gets considerably more freedom. It can receive an open-ended question, determine what information it needs, choose an approach and potentially take action on its own.
“I think agentic AI is really around a contained software system that’s allowed to work semi-autonomously,” Bilins said.
That autonomy is where things get interesting for advertising. An agent might eventually sift through proprietary data and supply signals, determine how to optimize toward different objectives and make adjustments without waiting for humans to analyze every report.
For anyone who has spent a career staring at advertising dashboards, this may sound less like technological disruption and more like an overdue workplace benefit.
Still, Bilins said the industry isn’t there yet. Many applications currently described as agentic AI involve pulling and analyzing reports or troubleshooting deals. True autonomous decision-making at scale remains further away.
“There’s a lot of infrastructure that needs to be put in place,” she said.
That includes reliable data, governance that determines what agents can and can’t do and enough transparency to understand their decisions. Systems also need to be auditable so buyers and sellers can develop confidence in allowing agents to do more over time.
Soon everyone gets an agent
Another complication is that there won’t necessarily be one AI agent politely running the advertising marketplace.
Buyers may have their agents. Sellers may have theirs. Each could arrive armed with different data, algorithms and KPIs while trying to optimize for different outcomes. After decades of humans disagreeing about what constitutes a successful campaign, machines may finally get their opportunity to disagree at computational speed.
“The beauty of it is because of the lower barrier to entry, you can have your own agent that optimizes for your particular outcomes,” Bilins said.
The beauty can quickly become complexity. If agents on both sides of a transaction are pursuing different goals, the industry needs governance and transparency around what each system is trying to accomplish.
It also needs common technical protocols.
“If all of the agents are trying to speak a completely different language, then it’s gonna be a mess to try to work with all the different partners in the ecosystem,” Bilins said.
That helps explain Magnite’s emphasis on openness and interoperability rather than betting everything on a single proprietary AI environment.
Adtech needs an AI common language
Bilins said trying to predict which AI technology ultimately wins is especially difficult because the field is moving so quickly. New models and development tools continue to lower the barrier to building software.
Rather than guessing where the technology lands, Magnite aims to support leading protocols, different models and standardized tools that can work across agents.
That approach also lets individual companies keep the data, algorithms and other intellectual property that make their systems distinctive.
“You should be able to bring your secret sauce to the table,” Bilins said.
Magnite can provide part of the infrastructure that lets companies put that secret sauce to work without requiring everyone in the marketplace to use the same recipe.
The result, Bilins argues, should be a healthier marketplace in which buyers, sellers and technology providers can develop their own AI capabilities while still communicating with one another.
Tedious work may disappear quietly
Some of AI’s biggest effects on advertising may ultimately seem surprisingly mundane.
Bilins predicts people may look back within a year or two and marvel at how much time they once spent performing tedious tasks that AI handles automatically.
Her analogy is the phone number. People once memorized numbers because they had to. Smartphones made the skill largely unnecessary. AI could make many routine advertising tasks disappear in much the same way.
“Oh my gosh, we spent so much time doing tedious tasks that now we don’t even realize AI is doing for us in the background,” Bilins said, describing how people may view today’s workflows in retrospect. “It’s just part of the way we work.”
That doesn’t mean humans can start clearing out their desks quite yet.
Brand safety is one area where Bilins expects human judgment to remain important. Brands still need to decide how they want to present themselves to consumers and establish the rules that AI systems operate within.
“AI is a machine,” Bilins said. “It does not have emotion or empathy.”
Even AI can’t kill the upfronts
And then there are the upfronts.
After surviving cable fragmentation, streaming, programmatic advertising, cord-cutting and countless predictions of their demise, television’s annual ritual of big negotiations apparently won’t be defeated by artificial intelligence either.
“I also think the upfronts aren’t gonna go away just because of AI,” Bilins said.
AI can help with planning and assist in negotiations. But Bilins expects the biggest deals between major brands and publishers to retain a substantial human component because they involve large commitments and significant risk.
So the machines may analyze the data, optimize campaigns and relieve people of hours of tedious work. They may even negotiate with other machines.
But when billions of advertising dollars are on the table, humans apparently still want another human sitting across from them.
For now, at least, nobody has invented an algorithm capable of replacing the upfront dinner.
You’re watching coverage from Beet Retreat Berkshires 2026. For more videos from this event, please visit this page.
AMENIA, N.Y. – Advertising technology has spent decades finding increasingly sophisticated ways to automate buying and selling media. Now artificial intelligence is promising to automate some of the work involved in operating all that automation.
Paige Bilins, senior vice president of product management at Magnite, sees potentially significant changes ahead as AI agents become capable of making decisions and acting on them. But she cautions that much of what the industry currently calls “agentic AI” is still closer to good old-fashioned automation with a shinier name.
“We’re still in very early innings around AI,” Bilins said in an interview with Beet.TV contributor David Kaplan at the Beet Retreat Berkshires.
Agentic AI isn’t just faster automation
Bilins draws an important distinction between AI automation and genuinely agentic systems.
Automation takes an existing task and performs it faster or more efficiently. An AI agent gets considerably more freedom. It can receive an open-ended question, determine what information it needs, choose an approach and potentially take action on its own.
“I think agentic AI is really around a contained software system that’s allowed to work semi-autonomously,” Bilins said.
That autonomy is where things get interesting for advertising. An agent might eventually sift through proprietary data and supply signals, determine how to optimize toward different objectives and make adjustments without waiting for humans to analyze every report.
For anyone who has spent a career staring at advertising dashboards, this may sound less like technological disruption and more like an overdue workplace benefit.
Still, Bilins said the industry isn’t there yet. Many applications currently described as agentic AI involve pulling and analyzing reports or troubleshooting deals. True autonomous decision-making at scale remains further away.
“There’s a lot of infrastructure that needs to be put in place,” she said.
That includes reliable data, governance that determines what agents can and can’t do and enough transparency to understand their decisions. Systems also need to be auditable so buyers and sellers can develop confidence in allowing agents to do more over time.
Soon everyone gets an agent
Another complication is that there won’t necessarily be one AI agent politely running the advertising marketplace.
Buyers may have their agents. Sellers may have theirs. Each could arrive armed with different data, algorithms and KPIs while trying to optimize for different outcomes. After decades of humans disagreeing about what constitutes a successful campaign, machines may finally get their opportunity to disagree at computational speed.
“The beauty of it is because of the lower barrier to entry, you can have your own agent that optimizes for your particular outcomes,” Bilins said.
The beauty can quickly become complexity. If agents on both sides of a transaction are pursuing different goals, the industry needs governance and transparency around what each system is trying to accomplish.
It also needs common technical protocols.
“If all of the agents are trying to speak a completely different language, then it’s gonna be a mess to try to work with all the different partners in the ecosystem,” Bilins said.
That helps explain Magnite’s emphasis on openness and interoperability rather than betting everything on a single proprietary AI environment.
Adtech needs an AI common language
Bilins said trying to predict which AI technology ultimately wins is especially difficult because the field is moving so quickly. New models and development tools continue to lower the barrier to building software.
Rather than guessing where the technology lands, Magnite aims to support leading protocols, different models and standardized tools that can work across agents.
That approach also lets individual companies keep the data, algorithms and other intellectual property that make their systems distinctive.
“You should be able to bring your secret sauce to the table,” Bilins said.
Magnite can provide part of the infrastructure that lets companies put that secret sauce to work without requiring everyone in the marketplace to use the same recipe.
The result, Bilins argues, should be a healthier marketplace in which buyers, sellers and technology providers can develop their own AI capabilities while still communicating with one another.
Tedious work may disappear quietly
Some of AI’s biggest effects on advertising may ultimately seem surprisingly mundane.
Bilins predicts people may look back within a year or two and marvel at how much time they once spent performing tedious tasks that AI handles automatically.
Her analogy is the phone number. People once memorized numbers because they had to. Smartphones made the skill largely unnecessary. AI could make many routine advertising tasks disappear in much the same way.
“Oh my gosh, we spent so much time doing tedious tasks that now we don’t even realize AI is doing for us in the background,” Bilins said, describing how people may view today’s workflows in retrospect. “It’s just part of the way we work.”
That doesn’t mean humans can start clearing out their desks quite yet.
Brand safety is one area where Bilins expects human judgment to remain important. Brands still need to decide how they want to present themselves to consumers and establish the rules that AI systems operate within.
“AI is a machine,” Bilins said. “It does not have emotion or empathy.”
Even AI can’t kill the upfronts
And then there are the upfronts.
After surviving cable fragmentation, streaming, programmatic advertising, cord-cutting and countless predictions of their demise, television’s annual ritual of big negotiations apparently won’t be defeated by artificial intelligence either.
“I also think the upfronts aren’t gonna go away just because of AI,” Bilins said.
AI can help with planning and assist in negotiations. But Bilins expects the biggest deals between major brands and publishers to retain a substantial human component because they involve large commitments and significant risk.
So the machines may analyze the data, optimize campaigns and relieve people of hours of tedious work. They may even negotiate with other machines.
But when billions of advertising dollars are on the table, humans apparently still want another human sitting across from them.
For now, at least, nobody has invented an algorithm capable of replacing the upfront dinner.
You’re watching coverage from Beet Retreat Berkshires 2026. For more videos from this event, please visit this page.
