Pragmatic Media Buyers View LLMs as Workflow Accelerators, Not Autonomous Ad Buyers

The hype around agentic AI at CES 2026 masked a deeper, more pragmatic reality for media buyers and marketers, revealing a critical tension between aspirational technology and immediate operational needs. While headlines screamed about autonomous decision-making, the conversations happening behind closed doors painted a picture of cautious adoption, emphasizing workflow automation over true transactional replacement. This divergence highlights a hidden consequence: the industry is being sold a future that demands significant investment and trust in systems not yet proven for high-stakes, real-time financial transactions. Those who can bridge this gap by focusing on tangible, immediate workflow improvements, rather than chasing the elusive autonomous agent, stand to gain a significant competitive advantage by solving current problems while others are still debating future possibilities.

The Pragmatic Divide: Why Agentic AI Isn't Buying Ads (Yet)

CES 2026 was awash in the promise of agentic AI, a future where intelligent agents autonomously manage complex tasks. Yet, beneath the surface of ambitious announcements, a stark contrast emerged between the industry's public pronouncements and the private anxieties of those tasked with allocating actual media dollars. The core thesis, as articulated by Sub Joseph, executive editor at Digiday, is that while large language models (LLMs) are undeniably reshaping workflows, they are not yet, and perhaps not soon, ready to replace human decision-making in critical areas like programmatic buying. This isn't a matter of technical limitations that will be solved next quarter; it's a fundamental mismatch between the probabilistic nature of LLMs and the deterministic, constraint-driven reality of programmatic advertising.

The immediate allure of AI, particularly agentic AI, is its potential to automate and accelerate. However, the conversations Joseph had at CES revealed a significant disconnect. Buyers, agencies, and even ad tech engineers largely agreed that LLMs can indeed speed up processes, but they are far from capable of autonomous buying. Programmatic buying, at its heart, relies on precise optimization within strict rules and constraints. LLMs, on the other hand, excel at probabilistic reasoning--making educated guesses based on vast datasets. Equating these two is akin to asking a brilliant poet to perform complex financial accounting; the underlying logic and required precision are fundamentally different.

"The consensus seemed to be that LLMs are not yet fit for autonomous buying... they excel at the probabilistic reasoning sort of side of things but... programmatic buying is built on deterministic optimization."

This technical gap, while narrowing, is compounded by deeper philosophical concerns. The idea of handing over ad dollars to AI agents, especially those trained on the notoriously "noisy, incomplete, and adversarial" programmatic supply chain, raises a red flag. The fear, articulated by many, is that instead of making advertising smarter, training AI on flawed data would simply amplify and accelerate existing blind spots. This is the essence of the "garbage in, garbage out" problem, a warning that resonated throughout many discussions. The programmatic ecosystem, with its inherent complexities and historical inefficiencies, is not a clean slate upon which to build autonomous decision-makers. Instead, it risks becoming a substrate that hardens existing flaws, making them more permanent and more expensive to rectify.

The NBC Universal announcement of enabling AI agents for buying traditional TV and digital, including a pilot program for an NFL game, stands in stark contrast to this prevailing sentiment. While this might represent a bold step forward for NBCU and its partners, it appears to be an outlier, a signal of a different strategic direction than what most buyers are currently comfortable with. Joseph notes that for such initiatives to be truly indicative of a market shift, there needs to be corresponding activity from the buy-side. Without that, it risks being a seller-driven narrative rather than a market-wide adoption. The distinction between workflow automation and true agentic transaction handling is crucial here, and many of the reported "agentic" activities are, upon closer inspection, more about streamlining internal processes than enabling autonomous ad buying.

"Teaching ai on that [programmatic supply chain] substrate would not make advertising necessarily smarter it would sort of make its blind spots more permanent and more like and faster and sort of like like more expensive."

The economic realities further temper the AI hype. While agencies like WPP and Omnicom are actively developing their "agent hubs" and next-generation platforms, much of this jockeying appears to be narrative-building rather than a reflection of widespread business shifts. The economic forecasts, which suggest a relatively stable market for consumer spending and advertising in the near term, mean that many CFOs are unlikely to approve major AI-driven overhauls. For non-advertising-funded businesses like P&G or Unilever, a pragmatic approach is paramount. They are not in a position to make drastic organizational changes based on speculative future capabilities. The cost of compute, including GPUs and token costs for LLMs, also remains a significant, often unclear, factor. As Wester Har of S4 Capital pointed out, the cost of inference is an active discussion point, and this will inevitably translate into how agencies charge clients, creating another layer of complexity and potential friction.

The conversation at CES was less about the far-flung future of autonomous agents and more about the immediate, tangible benefits of AI in workflow automation. This includes everything from planning and optimization to creative development. The real growth opportunity for marketing services businesses, according to Joseph, lies in mastering these workflow automation tools, differentiating through proprietary data and expertise. This pragmatic focus, while perhaps less glamorous than the agentic AI narrative, is what CMOs are actively seeking. They cannot afford major organizational shifts amidst current uncertainty, and the financial implications of AI are still too opaque to warrant large-scale adoption for transactional purposes.

"It feels as though this year will be the year where creator measurement kind of finds its footing... brands are no longer willing to sort of spend you know exorbitant amounts of money on on sort of creators without some without some robust sort of guardrails kind of there."

Beyond AI, the discussions around creator measurement also signal a shift towards pragmatism. As platforms begin to allow more data flow and brands demand clearer ROI, the industry is bracing for a year where creator measurement finds its footing. This implies a move away from unbridled spending towards more data-driven, accountable partnerships. This focus on tangible outcomes and verifiable results underscores the broader theme: the industry is prioritizing solutions that address current pain points and deliver demonstrable value, rather than chasing speculative technological advancements.

Key Action Items:

• Immediate Actions (Next 1-3 Months):

  • Identify and pilot workflow automation tools: Focus on AI applications that streamline planning, optimization, or creative processes, not transactional buying.
  • Develop clear "garbage in, garbage out" protocols: For any AI tool being considered, rigorously assess the quality and integrity of the data it will be trained on or operate with.
  • Engage in direct conversations with agency partners: Seek clarity on their AI strategies, focusing on how they plan to leverage AI for workflow efficiency rather than autonomous decision-making.
  • Investigate creator measurement solutions: Begin evaluating platforms and methodologies that offer robust, data-driven insights into creator campaign performance.

• Medium-Term Investments (Next 6-12 Months):

  • Build internal expertise in AI workflow integration: Train teams on how to effectively use AI tools to enhance existing processes, focusing on efficiency and cost savings.
  • Explore pilot programs for AI in non-transactional areas: Consider AI for tasks like audience segmentation, predictive analytics, or content personalization, where the stakes are lower than direct ad buying.
  • Establish clear metrics for AI-driven workflow improvements: Define how success will be measured for AI adoption, focusing on speed, cost reduction, and accuracy in operational tasks.

• Longer-Term Strategic Investments (12-18+ Months):

  • Develop a comprehensive AI strategy that accounts for compute costs: Factor in the ongoing expenses of AI inference, model training, and data infrastructure into long-term financial planning.
  • Monitor the evolution of LLM capabilities in deterministic environments: Keep abreast of technical advancements that might bridge the gap between probabilistic reasoning and the demands of programmatic buying, but remain cautious about premature adoption.
  • Advocate for industry standards in AI measurement and transparency: Contribute to discussions that aim to create reliable frameworks for evaluating AI's impact and ensuring accountability, particularly as transactional AI becomes more prevalent.