From Eyeballs to AI Agents: The New Economics of Marketing

Marketing has always operated around a simple proposition: Get in front of people. Capture attention, create preference and make it easy for the consumer to buy.

AI agents complicate every part of that formula.

Instead of opening a search engine, comparing products, reading reviews and navigating checkout, a consumer may increasingly tell an agent what they want and let software handle the rest. The agent can interpret preferences, compare options, check prices, evaluate constraints and eventually complete the transaction.

And if humans are no longer evaluating every option, companies may have to stop asking, “How do we get noticed?” and start asking, “How do we become the option an agent can confidently recommend?”

Members of the Senior Executive AI Think Tank—a curated group of experts specializing in machine learning, generative AI and enterprise AI applications—see that transition from different vantage points. Here, they explore what this shift could mean for companies, consumers and the future of commerce—and what businesses may need to rethink as AI agents take on a greater role in the buying process.

“When an agent is doing the shopping, the checkout experience becomes the product.”

– Anirban Poddar, Senior AI Product Manager at Meta

SHARE IT

Make the Transaction Layer the Product

Anirban Poddar, Senior AI Product Manager at Meta, believes the first thing companies will pay for is not exposure but the ability to complete a transaction cleanly.

“Companies will pay for the smoothest transaction layer,” Poddar says. “When an agent is doing the shopping, the checkout experience becomes the product—how cleanly a company’s payment stack connects, and how fast the agent can complete a purchase without the user ever touching a form.”

That puts pressure on everything between recommendation and payment: APIs, authentication, payment systems, inventory connections and checkout flows. Companies that make those connections easy for agents could become preferred destinations even when consumers never visit their websites.

Poddar also sees a second opportunity in what he calls “taste marketing for agents.” Rather than describing products only through conventional attributes such as price, size or specifications, brands will need rich qualitative information that helps an agent determine whether a product fits a particular consumer.

“The agent already understands the user—their taste, their constraints, what ‘good’ looks like for them,” he says. “So companies will need to present their products in a way the agent can match against that understanding.”

Preserve the Human Reasons to Shop

Not every part of commerce should become more efficient.

John Cho, Chief Information Officer at EdgeConneX, points to a dimension of shopping that algorithmic commerce could easily overlook: Shopping can be social.

“One thing to remember is that there is a social and relational aspect of shopping, especially with your friends and family,” Cho says. “Some inefficiency is actually part of the value.”

His point matters because companies could mistakenly treat every human interaction as friction to eliminate. Wandering through stores, debating choices, changing one’s mind and sharing reactions may be inefficient from a transaction perspective, but valuable from an experiential one.

Cho therefore sees a hybrid future rather than a fully delegated one.

“Companies need to reframe the paradigm shift from going purely agentic to some hybrid,” he says. “The distinction may not be analog versus digital, but participatory versus delegated.”

For executives, that suggests separating purchases people want optimized from experiences people want to participate in. The best strategy may not be to automate every customer journey but to determine where human participation is itself the product.

“Paying to influence matching can and will introduce dysfunction. And we’ve already seen that happen with the way the platforms behave.”

– Marc Massar, Founder of AURA Labs

SHARE IT

Protect the Integrity of the Match

Marc Massar, Founder of AURA Labs, is skeptical of simply recreating advertising inside agentic systems.

“It remains to be seen how the economics of product placement will work with machine audiences,” Massar says. “But, we can be sure that humans will attempt to game the system.”

If brands can pay to influence an agent’s recommendation, the marketplace could reproduce the same dysfunction associated with paid placement today—only with fewer obvious signals to consumers.

“Paying to influence matching can and will introduce dysfunction,” he says. “And we’ve already seen that happen with the way the platforms behave.”

Massar argues that agents could instead create markets in which matching is based more explicitly on fit between buyer preferences and product attributes. His broader goal is to give consumers greater control over the data and preferences used to make those matches.

“Hopefully, agents will return power to consumers and give them choices they wouldn’t otherwise have,” Massar says. “Maybe we’ll see a place where consumers finally have data sovereignty and choose what personal information they trade.”

Turn Operational Performance Into Marketing

Rishi Kumar, Chief Transformation Officer at Matchingfit, believes the next scarce commodity will be machine trust.

“I think companies will increasingly pay not for attention, but for the likelihood that an AI agent will trust and choose them,” he says.

That changes what counts as marketing infrastructure. Product data, pricing, fulfillment, warranties and returns are no longer merely operational details that matter after a purchase. They can determine whether the purchase happens at all.

“The bigger shift is that marketing and operations start to blur,” Kumar says. “A late shipment, misleading claim or painful return may influence the next recommendation more than an ad ever could.”

That means executives should treat fulfillment accuracy, return performance, product quality and customer-service outcomes as acquisition metrics, not just operating metrics.

“In agentic commerce, performance itself becomes marketing, and trust becomes something machines can measure,” Kumar says.

Pay for Inclusion, Not Impressions

Jim Liddle, Entrepreneur, Investor, Advisor and Enterprise AI Strategist, sees the advertising model moving from exposure to selection.

“I think we will see companies shift from buying attention to buying inclusion,” he says. “They will be paying to get into the agent’s shortlist, recommendation logic and transaction path.”

That distinction could fundamentally change how marketing budgets are allocated. An impression has value because it might influence a future decision. An agent recommendation is much closer to the decision itself.

“The main shift will be from impressions to recommendations, from reach to trust and ultimately from clicks to completed transactions,” Liddle says.

The consequences extend into technical infrastructure. Companies will need machine-readable product information, reliable feeds, agent-facing APIs and reputation systems that make it easier for automated systems to understand what a business offers.

“In this new world if the AI Agent can’t read you, you don’t exist,” Liddle says.

Be Prepared to Define ‘Best’

Sai Krishna Reddy Mudhiganti, Staff AI/ML Software Engineer at Samsung Semiconductor, sees a less obvious battleground emerging.

“As an AI agent builder myself, the next advertising battle may be over who gets to define ‘best,’” he says.

That could mean companies spending heavily on benchmarks, comparison engines, ratings and evaluation methodologies that shape what agents consider important.

“A laptop maker could sponsor a ranking that heavily rewards battery life because that is where its products win,” Mudhiganti says. “If an agent treats that ranking as neutral evidence, the sales pitch has already entered the decision.”

The strategic response is not simply to create more favorable scorecards. Executives should ask which measurements genuinely predict customer value—and make sure those measures are transparent enough that agents and consumers can distinguish evidence from promotion.

“An agent can check every fact and still inherit the seller’s definition of value,” Mudhiganti says.

Move Upstream From Intent to Value

Rodney Mason, Chief Marketing Officer at Minty, argues that the scarce resource will increasingly be intent.

“Companies will stop paying for eyeballs and start paying for intent,” he says.

New research from Northwestern University’s Retail Analytics Council and Minty provides an early signal of where those budgets could move. A survey of 150 senior commerce decision-makers found that marketers plan to direct 63 cents of every new e-commerce marketing dollar toward AI commerce tools and cashback and savings apps rather than traditional digital advertising. Additionally, 79% expect their primary customer to be AI-assisted in 2027.

Mason says the lesson is that value beats visibility.

“The winners will practice Proactive Commerce, reaching shoppers and their AI agents with total value before they search, exit or abandon, and recovering the invisible shoppers that ads never reach,” he says.

That does not necessarily mean abandoning marketing. It means changing when marketing happens and what it delivers.

“When AI agents evaluate and transact, commerce shifts from emotional persuasion to algorithmic utility.”

– Geetha Kumari Kommepalli, Founder, CTO and Chief AI Officer at Thriven Advisory

SHARE IT

Make Commerce Machine-Readable

Geetha Kumari Kommepalli, Founder, CTO and Chief AI Officer at Thriven Advisory, sees three forms of infrastructure becoming commercially valuable: access to context, deterministic machine interfaces and trust verification.

“When AI agents evaluate and transact, commerce shifts from emotional persuasion to algorithmic utility,” she says.

That means brands need to make their information accessible not only to consumers but also to the systems acting on their behalf.

“Companies will no longer pay for human attention,” Kommepalli says. “Instead, they will pay for context injection and retrieval placement, deterministic machine endpoints and algorithmic trust verification.”

The emerging infrastructure already points in this direction. Google’s Universal Cart, for example, represents an effort to make shopping and checkout executable across AI-driven interfaces rather than requiring consumers to move through conventional retailer journeys.

Kommepalli expects the consequences to extend beyond marketing budgets.

“The impact is seismic,” she says. “Visual brand equity will collapse for routine goods, replaced by machine-to-machine pricing arbitrage.”

That makes structured product information, real-time pricing, inventory access and verified claims strategic assets rather than technical housekeeping.

Make Customer Service Part of the Sales Pitch

Pon Murugesh Devendren, SAP Enterprise AI Architect at Deloitte, argues that agents may make the most mundane parts of the buying journey unexpectedly important.

“I think companies will pay to remove friction from the decision, not just to appear in the search results,” he says.

An agent evaluating two similar products can easily favor the one with a clearer warranty, dependable delivery date or return policy. In that environment, customer service is not merely post-sale support.

“That makes customer service part of the sales pitch,” Devendren says.

The same logic applies when something goes wrong. If an agent learns that a retailer repeatedly creates problems for the consumer, that experience can influence future choices.

Companies may therefore need to optimize for agent-readable service policies just as they optimize product feeds today.

“The risk is that agent platforms charge for preferred placement,” Devendren says, “turning a recommendation that feels personal into another paid promotion.”

Win at the Model Layer

Yogesh Malik, CEO of Way2Direct B.V., describes the fundamental change as a move from an “eyeball layer” to a “model layer.”

“Companies will pay for trust and relevance at the model layer rather than attention at the eyeball layer,” he says.

That means marketing increasingly becomes a data problem. A product must be able to explain itself in ways an AI system can evaluate: What is it? Who is it for? What does it cost? How does it compare? What constraints does it satisfy?

“When an AI agent evaluates options on a consumer’s behalf, the criteria it uses and the preferences it has learned become the purchase funnel,” Malik says.

The implication is that product specifications, metadata, reviews, availability and evidence of quality must work together as a coherent machine-readable representation of the product.

For executives, that suggests treating product data as a revenue asset. A catalog that is technically accurate but incomplete, ambiguous or difficult for models to interpret may become a competitive liability.

“The brands that will win are not the ones with the best ads,” Malik says. “They are the ones with the clearest, most structured, most trustworthy product data.”

How to Prepare for Agentic Commerce

• Make the transaction layer agent-ready. Build clean integrations between product data, inventory, payments and checkout so an agent can complete a purchase without unnecessary handoffs.
• Preserve human participation where it creates value. Identify which shopping experiences consumers enjoy precisely because they involve discovery, debate, social interaction or physical engagement.
• Protect the integrity of matching. Establish clear rules for paid placement and disclose commercial incentives so recommendations do not quietly become advertisements.
• Turn operational performance into a marketing metric. Measure fulfillment, returns, customer service and product accuracy as factors that can influence future AI recommendations.
• Optimize for recommendation inclusion rather than impressions. Build the structured data, reputation signals and infrastructure agents need to evaluate and select your products.
• Audit the definition of “best.” Review the benchmarks, rankings and comparison services influencing AI decisions and determine whether they measure genuine customer value.
• Invest in intent and value earlier in the funnel. Give consumers and their agents compelling combinations of price, savings, relevance and availability before traditional advertising would reach them.
• Make your product machine-readable. Provide structured, real-time information that can be retrieved, verified and acted on by AI systems.
• Treat customer service as part of acquisition. Make warranties, delivery commitments, returns and support policies clear enough for agents to compare before purchase.
• Build trust at the model layer. Treat product data, evidence, provenance and relevance as strategic assets that determine whether AI systems can confidently recommend your brand.

The New Currency of Commerce

The common thread across these perspectives is that companies may still pay for access to customers—but the definition of access is changing. In an agent-mediated marketplace, the valuable position may be inside the recommendation, embedded in the transaction path or represented by trusted data that an AI system can verify.

That does not necessarily mean advertising disappears. Instead, the economics of influence may move closer to the actual decision: inclusion, intent, trust, transaction readiness and measurable performance. The companies that prepare for that shift now will have an advantage not because they shout louder, but because their products give both humans and machines better reasons to choose them.