Georgia-Pacific’s decision to delay handing over programmatic buying to autonomous agents captures a growing tension in digital advertising between the allure of automation and the reality of a fragmented ecosystem. While many advertisers dream of slashing costs and removing human oversight, the consumer packaged goods giant argues that the underlying media supply chain remains too messy for agents to deliver genuine value. This stance reflects a broader industry debate: should brands chase the latest AI hype first, or fix the foundational plumbing before turning on the robots? By pausing, Georgia-Pacific signals that short‑term efficiency gains could be illusory if they simply accelerate waste rather than eliminate it.
The promise of buying agents—software that autonomously negotiates, bids, and optimizes ad purchases—rests on the assumption that repetitive tasks can be safely handed off to machines. In theory, this reduces labor, speeds up decision‑making, and uncovers hidden efficiencies. Yet Paras Shah, senior director of digital media at Georgia‑Pacific, warns that automation without quality inputs merely reproduces existing flaws at scale. An agent can only follow the rules it is given; if those rules are built on inflated impression counts, duplicate inventory, or opaque fees, the machine will optimize toward those flawed metrics faster than any human team could.
Shah’s analogy of “automating a repetitive task where waste continues to exist” cuts to the heart of the problem: programmatic still suffers from excessive middlemen, inconsistent data signals, and misaligned incentives between publishers, exchanges, and demand‑side platforms. Dropping an AI agent into this environment does not magically create transparency; it simply accelerates the same inefficient loops. Consequently, Georgia-Pacific has invested the past two years in a cleanup effort—pruning unnecessary supply‑side partners, applying pre‑bid inventory scoring, and tightening controls over which impressions are eligible for purchase.
This groundwork is not a side project but a prerequisite for any meaningful agent deployment. By reducing the number of SSPs in the flow, the company cuts down on duplicate bids and reduces latency. Pre‑bid scoring lets Georgia‑Pacific assess viewability, brand safety, and audience relevance before the auction even begins, ensuring that the inventory presented to any future agent meets a quality threshold. Tightening purchase rules—such as enforcing frequency caps and blocking known fraud domains—creates a cleaner decision‑space where an agent’s optimization logic can actually improve outcomes rather than merely chase cheap, low‑quality impressions.
When asked about the future, Shah struck a balanced tone: “Would we consider buying agents in the future? Yes we would but it’s more important that we try to get the best quality impression at the lowest possible cost, and then at that point we try to automate that repetitive task.” This sequential mindset treats automation as the final step in a process improvement loop, not the opening move. It mirrors the lean manufacturing principle of stabilizing a process before attempting to speed it up—a lesson that many ad tech vendors sometimes overlook in their rush to showcase AI capabilities.
The broader market reflects this split. At the Digiday Programmatic Marketing Summit last December, numerous brand‑side marketers echoed Shah’s caution, while holding companies like WPP and Omnicom, along with ad‑tech vendors, have been pushing aggressively toward agent‑to‑agent buying pilots. The former group worries about relinquishing control over media quality and brand safety; the latter sees agents as a way to scale operations and capture efficiencies across massive client portfolios. This divergence underscores differing risk tolerances and strategic timelines: brands protecting long‑term equity versus agencies chasing short‑term margin gains.
Ian Maxwell, CEO of Converge, summed up the skepticism succinctly: “There’s a tendency to confuse automation with intelligence. Agentic buying won’t magically improve media performance if it’s simply automating inefficient workflows, poor data or bad commercial decisions.” He points out that the real dysfunction in programmatic lies not in the workflow steps themselves but in the incentive structures that reward volume over value, clicks over conversions, and opaque arbitrage over transparent pricing. Until those incentives are realigned—through better contracts, clearer measurement, and shared accountability—automating decisions will merely encode the status quo into faster‑executing software.
Despite the hesitation, momentum behind buying agents is undeniable. Venture capital continues to flow into autonomous media‑buying startups, and major platforms are embedding agent‑like capabilities into their DSPs. However, widespread adoption will likely unfold in pockets rather than a wholesale shift. The risks—brand safety breaches, unintended bidding wars, and regulatory scrutiny—remain significant, while the upside is still unproven at scale. Moreover, the absence of universal standards for how one agent communicates with another (think of an “agent‑to‑agent” protocol akin to OpenRTB) creates friction that hinders interoperability and limits the potential for network effects.
Mateusz Jędrocha, chief product officer at Adlook, offers a provocative reframing: “The truth is the industry is thinking about integrating agents wrong… But the true value of AI agents isn’t in replicating what humans already do.” In other words, rather than using agents to automate existing bid‑shading or frequency‑capping logic, companies should explore novel applications—such as dynamic creative optimization driven by real‑time contextual signals, or cross‑channel budget reallocation based on predictive lifetime value models. When agents tackle tasks that are genuinely beyond human capacity, their impact can be transformative rather than merely incremental.
Evidence of this broader experimentation is already appearing. Publishers are testing autonomous tools to sell premium inventory directly to advertisers, hoping to bypass the complexity of traditional exchanges while offering greater transparency. Simultaneously, ad‑tech firms are expanding agent concepts into emerging channels like connected TV, digital out‑of‑home, and retail media networks, where inventory fragmentation and pricing opacity create fertile ground for AI‑driven discovery. These pilots suggest that the most promising use cases may lie outside the traditional display banner waterfall that has long defined programmatic.
Holding companies are also shifting their narrative from vague “transformation” buzzwords to concrete outcome‑based promises. Moloco, though not a pure‑play DSP, exemplifies this trend by leveraging its mobile app‑install expertise to expand into CTV and other video environments, building agent‑like optimization layers that predict install likelihood across disparate screens. Such moves illustrate how companies with strong data foundations in one domain can extend their AI capabilities laterally, provided they first ensure data quality and measurement consistency in the new territory.
For marketers navigating this landscape, the takeaway is clear: prioritize hygiene before heralding autonomy. Begin by auditing your supply path—identify duplicate SSPs, assess fee transparency, and verify that your inventory signals (viewability, brand safety, audience segmentation) are reliable and deduplicated. Invest in a pre‑bid scoring layer or partner with vendors that offer robust fraud and suitability filtering. Only once you have confidence that the impressions you are buying meet a high‑quality threshold should you consider piloting an buying agent in a limited, well‑defined scope—perhaps a single geographic market or a specific product line.
When you do run a pilot, define success metrics that go beyond CPM or CTR; look at incremental lift in brand‑affinity studies, lift in purchase intent, or reductions in wasted spend on non‑human traffic. Use the agent’s decision logs to audit whether it is genuinely optimizing toward your stated goals or merely exploiting loopholes in a broken system. Finally, maintain a human‑in‑the‑loop for oversight and exception handling, allowing the agent to handle repetitive, high‑volume tasks while strategists focus on creative strategy, audience insights, and long‑term brand building. By following this staged, quality‑first approach, you can harness the promise of AI agents without falling into the trap of automating waste.