The Digiday Podcast: "The Case Against AI Agents for Programmatic Ad Buying"

Date: December 9, 2025
Host: Tim Peterson (Executive Editor, Digiday Media)
Guest: Christopher Francia (Director of Product Development & Client Performance, Attention Arc)
Guest Co-host: Seb Joseph (Executive Editor of News, Digiday)

Overview

This episode delves into the present and future of AI agents in programmatic ad buying, specifically arguing why, despite the hype, AI (particularly large language model-driven agents) isn’t upending the space as quickly as some imagine. Live from the Digiday Programmatic Marketing Summit (DPMS), guest Christopher Francia of Attention Arc details the technological and practical limitations of current AI agents, while the Digiday team also digests major industry news (especially the landmark Netflix-Warner Bros. Discovery merger).

Key Discussion Points and Insights

1. State of AI Agents in Programmatic Buying (00:10–03:01, 34:54–55:41)

• Current Use of AI Agents

  • AI is automating “monotonous, mundane, labor-intensive” campaign tasks, not the core, high-value commercial elements.
    • “The more manual aspect of the job seems to be automated away.” – Seb Joseph (02:20)
  • “AI agents right now are good for stuff you would otherwise hand off to an intern.” – Tim Peterson (02:47)

• Technical and Strategic Limitations

  • AI agents, specifically large language models (LLMs), routinely “hallucinate” or produce unreliable outputs. This is particularly dangerous in “brand-safe” environments where a single mistake could damage hundred-year-old reputations. (35:24, 39:50)
  • LLMs are excellent for narrowly tailored, specific tasks. As complexity increases, so does unreliability. (35:08, 35:22)
  • Deterministic, classical machine-learning models (not LLMs) underpin real-world performance in Meta’s Advantage Plus and Google’s Performance Max. The “AI agent” narrative is often just marketing. (37:14–38:06)

• Industry Adoption and Definitions

  • The definition of “AI agents” is fuzzy and inconsistent.
    • “No one can agree on what an AI agent is…but the creative side…has had AI-generated creative for a bit now.” – Tim Peterson (38:56)
  • There's little to no current adoption of “agentic AI” for actual media buying among practitioners; most buy-side systems remain manual or rule-based. (46:50)
    • “Not a single hand.” – (on show-of-hands for actual AI-agent usage at DPMS) (46:52)

• Where AI Adds Value

  • AI is most productive in campaign ideation, insights, summarization, and repetitive, rule-based tasks like cloning campaign structures or parsing performance data. (51:32–54:41)
  • “AI is good for insights and summarization and helping you. It’s a tool helping you ideate and things like that…But they're in the driver's seat.” – Christopher Francia (51:44)

2. Major Industry News: Netflix Acquires Warner Bros. Discovery (04:38–24:27)

• Deal Overview

  • Netflix’s $83B acquisition (including assumption of debt), focusing on Warner Bros. Studios and streaming (HBO Max), not CNN or cable assets.
  • The speed and inevitability of the announcement, with Netflix even emailing subscribers before regulatory/due diligence was finished, drew attention.
    • “I don’t remember the last time a company sent this email to its customers so quickly after announcing an acquisition.” – Tim Peterson (10:54)

• Strategic Motivations

  • The acquisition is about “accumulating more data”—intellectual property in this context—mirroring a trend in ad holding companies (Omnicom-IPG).
    • “This is about intellectual property, about the film and TV library, which is data in its own kind of way.” – Tim Peterson (10:23)
  • Netflix’s own attempts to develop franchises organically (e.g., Roald Dahl library) have been expensive with limited payoff; this buy gives them core, proven brands (Harry Potter, DC, etc.).
    • “Netflix has probably kind of wasted a small sort of fortune trying to build its own IP.” – Seb Joseph (11:37)

• Potential Outcomes & Industry Impact

  • Theatrical model: Netflix may preserve tentpole theater releases for blockbusters but will likely use its newfound IP to entice both subscribers and advertisers.
    • “I can’t imagine them sort of putting the next Batman film…straight to [streaming]…these are $1 billion franchises.” – Seb Joseph (13:07)
  • Regulatory and competitive hurdles remain (potential DOJ intervention, Paramount’s hostile counter-offer, resistance from Hollywood, etc.). (14:06–16:27)
  • "This is far from over...It's really interesting to think what are going to be the impacts of this." – Tim Peterson (16:27)

3. AI Licensing, Lawsuits, and Industry Shifts (25:38–33:15)

• Meta and AI Content Licensing

  • Meta is now paying publishers (e.g., CNN, USA Today) to license content for AI model training, a new revenue source but with an uncertain duration.
    • “Meta has a history of saying, ‘Hey, publishers, we're open for business’…and then within a matter of years saying, ‘OK, we changed our mind.’” – Tim Peterson (25:38)
  • Microsoft and others are quickly following suit, while lawsuits (e.g., NYT vs. Perplexity/OpenAI) proliferate.

• OpenAI’s Competitive Position

  • OpenAI is under “Code Red,” shifting all focus to ChatGPT amid Google’s rapid Gemini 3 advancements, deprioritizing development of an ad product.
  • Unclear how or when a scalable ad revenue-share model for ChatGPT emerges, potentially ceding ground to Google’s faster ad integration and longstanding dominance in paid search/display.
    • “The longer [OpenAI] don't have ads…does that work against them...the more time Google has to lock in the market.” – Seb Joseph (30:08)
    • “Google definitely has the leg up. Google already has search dollars and display dollars…inertia is real in advertising.” – Tim Peterson (32:28)

4. Structural Barriers to AI Agents in Programmatic (34:54–55:41)

• Speed and Infrastructure Constraints

  • The technical reality of programmatic: bid requests have ~15–100ms windows; even optimized LLMs can’t process (parse, score, and respond) fast enough at systemic scale.
    • “When you take that, it becomes a physics problem…It’s not going to back out ever in terms of an ROI.” – Christopher Francia (41:38, 44:35)

• Industry Stagnation and Standards

  • Proposed standards and frameworks (Agentic RTB, User Context Protocol, Amazon’s RTB Fabric) aim to speed intra-platform data transfer and communication, but none solve LLM latency at real scale.
  • Large tech platforms (Google, Amazon, Trade Desk) have little interest in protocols standardizing or opening budget-shifting among DSPs—they have no incentive to level the playing field for smaller rivals.
    • “…That's not good for Google's business model. It's definitely not good for Trade Desk's…and it's not good for Amazon...It's really good for a mid-tier DSP who's just looking for any chance to prove itself.” – Christopher Francia (47:51)

• Fragmentation is Baked In

  • Agencies like Attention Arc build to platforms with the biggest budgets, not smaller, innovative APIs.
    • “We are not spending our resources and our time building to a small API…we have to go where most of the budget is.” – Christopher Francia (49:24)

• Where AI Works Best

  • Repetitive, well-defined, lower-value tasks; rapid ideation and insights; creative workflows (not real-time dynamic creative modifications, due to brand risk).
  • “We are experimenting with other stuff, but we're really more into the deterministic ML stuff when it comes to optimizations.” – Christopher Francia (54:41)

Notable Quotes & Moments (with Timestamps)

• On AI "Interns":
  • “These AI agents right now are good for stuff that you would otherwise hand off to an intern.”
    – Tim Peterson (02:47)
• On LLM Hallucinations & Real-World Limits:
  • “It has to do very specific, narrowly tailored tasks. The more complicated a task becomes, the harder it is for AI to figure out what it’s supposed to be doing.”
    – Christopher Francia (35:08)
  • "[AI is] not going to be trusted on activation... brands really want to control their image and their reputation and where they're running and what their creatives say. Taking that out sounds like a great idea…but that generative real-time [creative] is very dangerous until you can have full control, which we're just not there yet."
    – Christopher Francia (39:50)
• Adoption Reality Check:
  • On asking a live audience if they use agentic AI for ad buys:
    • “Not a single hand.”
      – Tim Peterson (46:52)
• On Big Tech Resistance to Standards:
  • “That's not good for Google's business model...Trade Desk...Amazon…It’s really good for a mid-tier DSP who’s just looking for any chance to prove itself.”
    – Christopher Francia (47:51)
• On Brand Risk and Practical AI:
  • “It just takes one mistake for that brand’s 100 year legacy to have a huge crisis.”
    – Christopher Francia (39:50)
• On AI in Creative vs. Media Buying:
  • “It’s been the most disruptive in creative workflows… the resources you need to do iterations of creatives have now reduced, which has made it for us to do more, take on more clients.”
    – Christopher Francia (38:06)
• On Netflix’s Deal-Making Pace:
  • “Netflix then announces the next morning, hey, we're acquiring Warner Brothers Discovery, everybody...This happened so quickly.”
    – Tim Peterson (06:56)
  • “This feels like Netflix kind of preempting that by, like, if everyone accepts this as reality, then it’s going to be inevitable that it'll close.”
    – Tim Peterson (10:54)
• On the ROI of LLMs in Bidstreams:
  • "The cost of doing that, to make it that fast, is not going to back out ever in terms of an ROI."
    – Christopher Francia (41:38)

Timestamps for Major Segments

• 00:10–03:01: Setting the stage: Role of AI in programmatic today, DPMS recap.
• 04:38–24:27: Deep-dive into Netflix’s acquisition of Warner Bros. Discovery: motivations, implications, industry parallels.
• 25:38–33:15: Rapid-fire industry news: Meta, Microsoft, legal action over AI content licenses; OpenAI v. Google in ads.
• 34:54–55:41: Interview with Christopher Francia: breaking down the technical, business, and operational limits of current AI agents in programmatic ad buying.

Tone & Style

The conversation is candid, nuanced, and skeptical—with a clear eye toward both the real technical hurdles and the ever-present marketing spin around ad tech and AI. There’s a healthy skepticism of overblown AI promises, paired with practical optimism about incremental gains in agency workflow, ideation, and performance reporting.

Summary

This episode makes a compelling case that, despite the hype, large language model-based AI agents are nowhere close to autonomously running programmatic ad buying. Real impact lies in automating rote work, enriching campaign insights, and accelerating creative iteration—not in replacing high-value strategic or commercial jobs. The biggest barriers are not just technological (speed, reliability, interpretability) but also business incentives and industry inertia. Meanwhile, seismic shifts continue at the platform and publisher level, promising even bigger changes ahead for media and advertising in 2026.