Large language models have infiltrated nearly every corner of digital advertising except the place that matters most: the moment when money actually changes hands. The distinction is not semantic. Planning tools, creative optimization, campaign setup, and reporting dashboards all welcome LLM integration. Platforms deploy agentic capabilities for workflow automation. But autonomous spending authority? That switch remains conspicuously off.

Jeffrey Hirsch delivered the reality check at Consumer Electronics Show in Las Vegas last week. His ad tech trading automation platform QuantumPath can compress two-hour workflows into 10-minute executions with near-zero errors, according to Hirsch. Yet the platform deliberately stops short of actually deploying budgets. "At QuantumPath, we want to automate the workflow, not the buying decisions," according to Hirsch, the company's CEO.

The sentiment reflects prevailing industry posture rather than contrarian positioning, according to reporting published January 14 by Digiday. Across agencies, platforms, and infrastructure providers, an invisible boundary separates automation that accelerates human work from automation that replaces humans at the point of spend.

Some of that caution stems from institutional self-preservation. But resistance runs deeper into programmatic advertising's technical foundations and unresolved accountability questions when machines mishandle inventory, according to the analysis.

Technical mismatch between LLMs and auctions

The most fundamental barrier involves architectural incompatibility. Large language models operate in open-ended semantic space, sampling probabilistically to generate outputs. Programmatic auctions demand fast, repeatable, deterministic logic.

That mismatch keeps LLMs parked at transaction edges—planning, setup, reporting, and analysis—rather than at the bidding core itself, according to industry executives interviewed for the report.

The boundary is not being treated as permanent, however. Michael Richardson characterized decision-level autonomy less as a technical limitation and more as an engineering timeline. Richardson, vice president of product at Index Exchange, expects more advanced autonomy to move out of experimentation as computing becomes cheaper and infrastructure matures. "It's not going to be broadly deployed" yet because of cost, readiness, and unresolved use cases, according to Richardson.

The models will eventually get there. The harder question for advertising executives is whether they should.

Data quality creates unreliable foundation

"That's the big concern for me: unreliable inputs produce unreliable decisions," according to Tom Swierczewski, vice president of media investment at Goodway Group. "For LLMs to buy autonomously in programmatic media, they'd need bidstream data—and that data is deeply flawed."

Advertising's modern data foundation remains riddled with distortions. Last-click bias, siloed walled gardens, platform-reported metrics that resist auditing, and persistent lack of incrementality adjustment continue shaping how performance gets measured, according to Swierczewski.

Training autonomous systems primarily on those inputs does not make systems smarter. It scales their blind spots. Because learning occurs continuously, those distortions do not merely persist—they become self-reinforcing, according to the analysis.

"The industry needs AI to manage complexity and move faster," according to Paul Boruta, CEO and founder of ad tech platform Slingwave. "But it should not hand that intelligence to systems that are optimizing toward the wrong signal."

That explains why so much current LLM investment focuses on plumbing rather than pilots, according to Digiday's reporting. Platforms modernize infrastructure, containerize auctions, open APIs, and lower technology switching costs—all prerequisites for autonomy—while deliberately keeping the bidder itself grounded in the same narrow, rule-bound machine learning that clears markets today.

Platform boundaries keep bidding deterministic

Yahoo DSP welcomes LLMs into orchestration layers and interfaces while keeping the core bidder rooted in deterministic bidding logic, according to Adam Roodman, general manager at Yahoo DSP. LLMs may drive dashboards and workflows. They are not being positioned as the engine deciding what to buy, when to buy it, or how much to pay.

"Nothing that we're doing at the moment would suggest that agentic or an LLM will take the place of bidding logic," according to Roodman. "I mean there could be parts of it eventually but at its core it will still be machine learning."

Yahoo launched agentic capabilities on January 6 enabling advertisers to automate campaign setup, troubleshooting, and optimization through natural language. The system operates under a "Yours, Mine, and Ours" framework allowing advertisers to bring their own AI models, use Yahoo DSP native agents, or connect both through secure interoperable protocols.

Even the most bullish LLM builders are recalibrating how they describe what their systems actually do versus what their branding suggests, according to Digiday. PubMatic's work with independent agency Butler/Till serves as a case study.

The companies positioned the effort as an end-to-end "agentic" campaign. Directionally, that framing holds. Operationally, it flattens important nuances, according to the report.

Butler/Till used an agent built on Claude to translate a human-written brief into a structured media plan. The resulting plan was passed to PubMatic, whose own AI systems mapped the intent to inventory, channels, and audience segments within its platform. Final parameters were reviewed and approved by Butler/Till staff before launch.

"We're intentionally being cautious on what we're directly and entirely attributing to agentic systems at this stage," according to Nishant Khatri, executive vice president of product management at PubMatic. "As the campaign continues, we expect greater clarity into efficiency and performance trends. Directionally, these results align with what we would expect from an early agentic campaign operating at a national scale."

PubMatic launched AgenticOS on January 5, positioning the infrastructure as the first operating system built specifically for autonomous advertising execution across premium digital environments. The platform includes containerized orchestration, agentic application layers, and transaction systems connecting decisioning directly to buying infrastructure.

What LLMs in advertising actually look like

The transformation underway is quieter than industry rhetoric suggests—labor compression, infrastructure rewiring, slow shifts in power across the advertising stack, according to Digiday's analysis.

The industry is not waiting for smarter machines. It is deciding who controls the machine that controls the money. Until that fight settles, LLMs can draft plans, build workflows, and run dashboards.

They just will not be handed the keys, according to the report.

The boundary between workflow automation and spending authority reflects fundamental questions about measurement accuracy, data quality, and accountability when autonomous systems make purchase decisions at scale. Industry infrastructure continues developing with standardized specifications for deploying containerized agents within real-time bidding infrastructure.

IAB Tech Lab released its Agentic RTB Framework version 1.0 for public comment on November 13, 2025, introducing standardized specifications for deploying containerized agents within real-time bidding infrastructure. The framework entered a public comment period extending through January 15, 2026.

The specification defines requirements for implementing agent services that operate within host platforms, leveraging containers deployed into infrastructure to enable delegation of bidstream processing tasks. According to the framework documentation, the model provides "minimal cost, latency and operational impacts" while establishing standard requirements for container runtime behavior.

Multiple protocols emerged during fall 2025 creating coordination challenges as platforms pursued different technical approaches. Six companies launched Ad Context Protocol on October 15, though skepticism emerged about whether another protocol was needed for agentic AI advertising automation.

Infrastructure versus execution authority

The distinction between building infrastructure and granting execution authority matters for understanding where LLM adoption actually stands versus where marketing materials suggest it has reached.

Amazon launched Ads Agent on November 11 at its annual unBoxed conference. The artificial intelligence agent automates campaign management tasks across Amazon Marketing Cloud and Multimedia Solutions with Amazon DSP. Campaigns only launch after advertisers review and approve automated recommendations, maintaining human oversight of advertising decisions.

Google announced on November 12 that Ads Advisor and Analytics Advisor would reach all English-language accounts across Google Ads and Google Analytics in early December. The tools represent Google's implementation of agentic conversational experiences powered by Gemini models, designed to accelerate data analysis and campaign management.

The platforms position these capabilities as "agentic" while maintaining guardrails that prevent autonomous spending without explicit human approval at decision points.

Adam Roodman at Yahoo DSP framed the development as fundamental workflow transformation. "Agentic AI changes how media buying actually gets done," according to Roodman. "By building it directly into Yahoo DSP and allowing advertisers to connect their own AI alongside ours, we're giving teams a faster, more flexible way to plan, optimize, and act, without sacrificing transparency or control."

The emphasis on transparency and control signals where platforms draw boundaries. Workflows accelerate. Transparency mechanisms expand. Control remains with humans at the spending moment.

Measurement infrastructure complications

The measurement challenges extend beyond data quality into fundamental questions about what gets measured and how attribution systems actually function.

Newton Research launched agentic AI analytics integration with Snowflake Cortex AI on November 4, 2025, enabling brands to run media mix modeling and incrementality analysis directly within secure data environments. The collaboration addresses accessibility barriers in marketing measurement by allowing customers to run Newton Research agents without data transfer requirements.

The Institute of Practitioners in Advertising released comprehensive guidance in March 2025 emphasizing that combining multiple measurement approaches delivers the most accurate picture of advertising performance. According to the IPA report, marketing mix modeling excels at understanding long-term effects and providing holistic views that disentangle media interactions.

Kochava research announced in September 2025 demonstrated that marketing mix modeling revealed TikTok campaigns generated an average of 35% higher incremental impact compared to last-touch attribution reporting.

The gap between what last-touch attribution reports and what incrementality analysis reveals illustrates why executives express concern about training LLMs on flawed measurement signals. Systems optimizing toward last-click attribution will systematically undervalue channels driving incremental lift.

Industry coordination around standards

The technical infrastructure necessary for autonomous advertising systems continues developing through industry coordination efforts attempting to prevent fragmentation.

IAB Tech Lab announced on January 6 a comprehensive agentic roadmap designed to scale artificial intelligence agent deployment across digital advertising without fragmenting the ecosystem through multiple incompatible protocols. The roadmap extends established industry standards including OpenRTB, AdCOM, and VAST with modern execution protocols rather than introducing entirely new technical frameworks.

Anthony Katsur, chief executive officer at IAB Tech Lab, stated in the January 6 announcement that the organization will make a significant engineering investment focused solely on artificial intelligence development. "Agentic execution is already part of how digital advertising operates today," according to Katsur. "Open, interoperable standards are what make that possible, and our focus is on scaling it responsibly."

The statement directly addresses mounting industry concerns about protocol proliferation. Multiple competing frameworks emerged during 2025, including the Ad Context Protocol launched October 15 with six founding members, and various proprietary implementations from major platforms.

Ray Ghanbari, chief technology officer at Index Exchange, emphasized ecosystem interoperability. "Our multi-party digital advertising ecosystem operates best when we all embrace, improve, and extend the transparency and interoperability standards that enable our industry," according to Ghanbari. "Tech Lab's Agentic Roadmap builds on this proven model, extending transparency and interoperability to service-to-agent and agentic transactions as well."

Autonomous systems pose platform business model questions

The statements reflect recognition that agentic capabilities pose existential questions for traditional programmatic platform business models.

Analysis published July 21, 2025, by Ari Paparo, founder and chief executive officer of Marketecture Media, argued that autonomous AI systems could automate campaign setup, targeting, and optimization functions currently handled by demand-side platforms, potentially eliminating the centralized role traditionally occupied by those platforms.

Industry veterans positioned Ad Context Protocol as different from real-time bidding infrastructure during January 2026 discussions, characterizing the emerging standard as "a protocol for investing" rather than the "day trading" approach embodied by OpenRTB.

Benjamin Masse, chief product officer at Triton Digital, articulated a distinction that challenges assumptions about how advertising technology should facilitate media transactions. The comparison draws directly from financial markets structure. Real-time bidding mirrors quantitative trading strategies that optimize individual transaction execution through algorithmic precision. Portfolio management operates at a higher strategic level, making allocation decisions across multiple investment vehicles based on long-term return objectives.

Brian O'Kelley elaborated this framework in a January 11 article, arguing that advertising faces allocation challenges rather than purely valuation problems. Portfolio managers concern themselves with how much capital to deploy across different opportunities, not just what individual assets are worth.

The distinction matters for understanding where LLMs might eventually gain spending authority. Portfolio-level allocation decisions differ fundamentally from impression-level bidding decisions in latency requirements, data needs, and accountability frameworks.

Consumer behavior shifts increase urgency

The infrastructure development occurs against backdrop of shifting consumer behavior that increases urgency around effective AI implementation.

Equativ's October 2025 survey of 4,000 North American and European consumers revealed 67% use AI more than once per week. The research found 38% search less frequently and 30% visit fewer websites as large language models reshape information discovery patterns.

The findings carry implications for brand visibility strategies. Curt Larson, chief innovation officer at Equativ, indicated that brands need to approach AI as a new channel requiring distinct considerations. Ensuring accurate product representation in large language models requires clear website structure and accurate content that machines can parse effectively.

A new advertising model proposed by Perplexity AI founder Aravind Srinivas outlined radical transformation where artificial intelligence agents, rather than humans, could become the primary target for advertisements. According to Srinivas in an interview conducted December 30, 2024, the future of advertising could involve AI agents acting as intermediaries between brands and consumers.

The proposed system would function differently from current digital advertising platforms. Multiple travel companies, platforms, and airlines would compete for the AI agent's attention in the backend, without the human user ever seeing traditional advertisements.

Industry responses to that vision have been mixed. Some technology experts raised concerns about potential conflicts of interest and the need for transparency in agent-based decision making. Others questioned whether consumers would be willing to delegate significant purchasing decisions to AI systems.

Where adoption actually stands

The gap between infrastructure development and autonomous spending authority reflects where the industry actually stands versus where aspirational roadmaps suggest it should reach.

McKinsey's Technology Trends Outlook 2025 report published in July 2025 identified 13 frontier technologies that could fundamentally reshape marketing strategies and advertising operations. The analysis positioned agentic AI as the most significant emerging trend for marketing organizations.

According to McKinsey's analysis, agentic AI represents a shift from chatbot interactions to virtual coworkers that can independently manage complex workflows. The research identified $1.1 billion in equity investment flowing into agentic AI technology during 2024. Job postings related to agentic AI increased 985% from 2023 to 2024.

Google Cloud survey data from April 2025 indicated that 88% of early adopter organizations implementing AI agents report positive return on investment across multiple business applications. The research found 52% of organizations using generative AI also leverage AI agents in production environments.

The adoption metrics measure workflow automation and decision support rather than autonomous spending authority. Organizations deploy agents to accelerate planning, optimize creative, and analyze performance. They do not grant agents authority to commit budgets without human approval.

The distinction between workflow acceleration and spending authority will likely persist until measurement accuracy improves, data quality issues get resolved, and accountability frameworks establish clear responsibility when autonomous systems make suboptimal purchase decisions.

For now, LLMs remain welcome everywhere in advertising except where the money actually moves. That is not an accident. It is deliberate industry positioning until fundamental questions about measurement, accountability, and control get answered satisfactorily.

Timeline

Summary

Who: Advertising technology executives including Jeffrey Hirsch (CEO, QuantumPath), Michael Richardson (VP of Product, Index Exchange), Tom Swierczewski (VP of Media Investment, Goodway Group), Paul Boruta (CEO, Slingwave), Adam Roodman (General Manager, Yahoo DSP), and Nishant Khatri (EVP of Product Management, PubMatic) discussed industry positioning on autonomous AI spending authority.

What: Large language models are being integrated into advertising workflow automation—planning, setup, reporting, and analysis—but platforms deliberately prevent LLMs from autonomous bidding decisions and budget deployment without explicit human approval at decision points. The industry draws a clear boundary between workflow acceleration and spending authority.

When: The analysis was published January 14, 2026, by Digiday following Consumer Electronics Show in Las Vegas where QuantumPath demonstrated workflow automation capabilities while explicitly stopping short of autonomous spending. The positioning reflects industry-wide approach developing throughout fall 2025 and early 2026.

Where: The boundary between workflow automation and spending authority exists across agencies, platforms, and infrastructure providers globally. Major platforms including Yahoo DSP, PubMatic, Amazon Ads, and Google Ads all implement agentic capabilities for workflow optimization while maintaining human oversight at the point of spend.

Why: Technical, measurement, and accountability challenges prevent autonomous spending authority. LLMs operate probabilistically, mismatching programmatic auctions' deterministic requirements. Advertising's data foundation contains distortions including last-click bias, siloed walled gardens, and lack of incrementality adjustment. Training autonomous systems on flawed inputs scales blind spots rather than intelligence. Industry executives express concern about unreliable measurements producing unreliable decisions, particularly when systems optimize toward wrong signals. Until measurement accuracy improves, data quality issues resolve, and accountability frameworks establish clear responsibility for autonomous purchase decisions, platforms will maintain human approval requirements at spending moments.

Share this article
The link has been copied!