Shelly Palmer, the Syracuse University media professor who runs the consultancy The Palmer Group, argued in an essay dated November 1, 2026 that "agentic marketing" has no settled meaning, and that the vendors selling it disagree on the question that carries the most risk: whether a human signs off before an AI agent's work goes live. The essay, "What Agentic Marketing Actually Means," proposes four questions to sort the pitches marketers will hear during 2027 planning - who decides the steps, who approves the actions, whether the system learns, and whom the agent works for - and argues that the answers will force companies to redraw their org charts. Two days later, on November 3, Anthony Katsur, chief executive of IAB Tech Lab, the advertising industry's technical standards body, called it a "Terrific read" on X and added that the Tech Lab also sees "APIs being relabeled as agents." The essay is an opinion piece. Its only market figures are Gartner estimates published in June 2025, and it tests no product against its own questions.

In Short

Lots of companies now sell marketing software they call "agentic," but they mean different things by the word, and some products sold as AI agents are older chatbots or automation tools with a new name. A tech commentator wrote that the biggest difference between these products is whether a person has to approve what the AI does before it reaches customers, and the head of the ad industry's main technical standards group publicly praised the essay. The essay says four questions - who plans the steps, who approves, whether it learns, and who it works for - show what a buyer is actually getting, and that companies will need new org charts that include the software.

What Palmer published, and where

The essay appeared on shellypalmer.com under the date November 1, 2026, a Sunday. Palmer promoted it on X on November 2 as "my Sunday essay," writing that it explores "the conflicting definitions as well as various ways org charts are going to change." The post opened with the line that frames the whole piece: "'Agentic marketing' currently means whatever the vendor selling it needs it to mean."

According to the biography attached to the essay, Palmer is Professor of Advanced Media in Residence at Syracuse University's S.I. Newhouse School of Public Communications and chief executive of The Palmer Group, "a consulting practice that helps Fortune 500 companies with AI strategy, implementation and governance, as well as technology, media and marketing."

An author's note states that the piece is not sponsored and that neither Palmer nor the company is paid for it, adding: "This work was created with the assistance of various generative AI models." It is not Palmer's first warning on the subject. At the ANA Masters of Marketing conference in Orlando last year, Palmer told marketers to be wary of AI agents and described them as synthetic employees.

Seven vendors, one word

The essay opens by naming the companies whose marketing uses the term. "Every major marketing platform is now selling 'agentic marketing,' and their definitions disagree on the questions that matter most: what counts as an agent, how much it decides on its own, and who approves its work," Palmer wrote.

Four platforms come first. According to Palmer, Salesforce "presents its marketing stack as Agentforce Marketing, built around autonomous agents that plan, create, personalize, and optimize campaigns." Adobe "ships Agent Orchestrator, a reasoning layer that sequences purpose-built marketing agents inside its Experience Platform." HubSpot "built its latest release cycle around content and nurture agents running a continuous plan-create-measure loop." And Amazon "renamed its unified ad platform Amazon Ads Agent in September, and its agentic AI guide repeats a prediction that autonomous agents will replace campaign-led marketing models by the end of the year."

Three more appear in a section titled "Same Phrase, Several Meanings." Bloomreach, in Palmer's account, defines agentic marketing as giving a system "a goal instead of a script." Klaviyo's definition "keeps a human approving everything that goes live." Amazon's guide "writes continuous human oversight out of the definition: autonomous systems that independently plan, execute, and optimize." Profound's definition "has the agent taking every step on its own, while the launch copy for Profound's own product promises marketers control of every approval."

Palmer's conclusion from the comparison is blunt. "The industry cannot agree on the single most consequential operational question in the category, whether a human signs off before anything ships, and some vendors cannot agree with themselves."

Adobe's product is more than a year old: the company introduced Agent Orchestrator for Adobe Experience Platform on September 10, 2025 with six agents, from audiences and customer journeys to site optimisation. Salesforce, in September 2026, gave its agents a runtime designed to pursue goals over weeks rather than single chats.

Amazon's record sits on both sides of the approval line Palmer draws. When the company introduced Ads Agent at unBoxed 2025 on November 11, 2025, campaigns launched only after advertisers reviewed and approved its automated recommendations. On September 29, 2026, Amazon folded its sponsored ads and its DSP into a platform branded Amazon Ads Agent, retiring the Amazon DSP name, and PPC Land reported on October 29 that the conversational agent itself was still in closed beta for eligible advertisers in 25 countries, leaving agencies uncertain what the rename covered. Palmer's essay mentions only the rename and the guide.

Agent washing, sixteen months on

The essay borrows its sharpest term from Gartner. "Gartner calls it 'agent washing.' The not-so-subtle art of sales organizations renaming chatbots, copilots, and robotic process automation as agents," Palmer wrote. "Of the thousands of vendors claiming agentic capabilities, Gartner estimated roughly 130 offer the genuine article, and it predicts that more than 40 percent of agentic AI projects will be scrapped by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls."

Those numbers are not new. Palmer links them to a Gartner press release dated June 25, 2025, and PPC Land reported both the 40% cancellation forecast and the estimate of about 130 genuine vendors in July 2025. The essay presents them in the present tense and does not say whether Gartner has revised either figure in the sixteen months since.

Later surveys point in the same direction without confirming Gartner's numbers. In August 2026, PPC Land reported that a KPMG survey of 2,145 senior leaders found 49% had scaled back, narrowed, delayed or paused AI agent rollouts after costs began to outweigh value, with only 7% reporting established return on investment. Advertising spending routed through agents remains small as well: Magnite's chief executive expects 2027 agentic ad spend to stay near $700 million at most.

Where the word comes from

A section headed "Agents Have Agency" traces the vocabulary. According to Palmer, the AI industry borrowed "agentic" from psychology, where Albert Bandura used it to describe human agency: "the capacity to make things happen intentionally, through your own actions." "The standard AI textbook has defined an agent for three decades as anything that perceives its environment and acts on it, a definition loose enough to cover a thermostat." The link points to the site for Stuart Russell and Peter Norvig's Artificial Intelligence: A Modern Approach.

Modern definitions, Palmer argued, add "goals, complexity, and limited supervision." Two AI developers are cited. OpenAI, in a paper on governing agentic AI systems, "treats agenticness as a spectrum: the degree to which a system can pursue complex goals in complex environments with limited supervision." Anthropic, in its engineering guidance on building effective agents, "draws the engineering line between workflows, where predefined code paths orchestrate every step, and agents, which decide their own processes and tool usage while the work is running."

Anthropic, the essay adds, "files both under one umbrella, agentic systems, and recommends the simpler workflow whenever the task is well defined." A model developer advising less autonomy where the task allows it sits some distance from vendor definitions that make autonomy the selling point.

One phrase, three relationships

The confusion, in Palmer's telling, runs deeper than vendors stretching a definition, because the same two words describe opposite relationships. In the first, the agent works for the marketer. In the second, which Palmer calls marketing to agents, the phrase "means optimizing your brand, your content, and your product data for the AI assistants that research, recommend, and increasingly buy on behalf of your customers." In search marketing, much of that work now goes under the name GEO. The second relationship already brings commercial friction: in September, Amazon said it had blocked Meta's Muse agent from buying on Amazon.com.

The third relationship belongs to the advertising trade. "The ad industry is already building a third relationship: IAB Tech Lab's AAMP protocols let buyer agents and seller agents plan, negotiate, and execute ad transactions with each other," Palmer wrote.

That description tracks the history of AAMP itself. IAB Tech Lab gave its agentic initiative that name on February 26, 2026, with an agent registry under its Trust and Transparency pillar. Buyer agents act for advertisers and agencies, seller agents for publishers. The AAMP 3.0 draft, published in September, is still open: on October 23 the Tech Lab extended the comment period to November 10, 2026. A rival, the Ad Context Protocol, launched on October 15, 2025, and in August IAB Tech Lab's chief operating officer, Shailley Singh, listed 13 business functions both address. A month later, Adform's chief technology and product officer said very little money runs through either.

"The vocabulary will not converge in time to help you," the essay states. Palmer's recommendation is that companies write their own definitions down and require everyone who uses the word, vendors included, to use them.

The four questions

The core of the essay is a test. "Four questions separate these meanings and sort every 'agentic' pitch you will hear during 2027 planning," Palmer wrote.

Who decides the steps?

"If a human designed the decision tree, the software is executing a plan, and an LLM writing the subject lines does not change that. An agent builds the plan itself, at runtime, from the goal and the data in front of it. This is the line between automation and agency."

On that test, much of what the ad industry currently runs would count as automation, some of it by design. Fluency, which manages about $3 billion in budgets, lets only deterministic agents act on live ad spend, and in January ad industry executives said they would not let large language models make bidding and budget decisions.

Who approves the actions?

"There are three honest answers: a human approves everything before it ships, a human approves exceptions above a threshold, or nothing requires approval. Each is a legitimate product with a completely different risk profile, compliance burden, and staffing model."

Few organisations have chosen the third. A Nylas survey of 1,026 developers and technology leaders, reported in March, found that only 4% let AI agents act without any human approval.

Does it learn?

"A system that runs the same play forever is a script with good handwriting," Palmer wrote. "A genuine agent closes a loop: it acts, measures the result, and changes its next action because of what happened." The essay asks buyers to establish where the loop closes, what data feeds it, and who can inspect the changes the system makes to itself.

Who does the agent work for?

This is the question with the widest commercial reach. "In one camp, the agent works for you: a digital employee executing your marketing. In the second, it works for your customer, researching and choosing among brands using criteria you will never see in a dashboard." Then comes the sentence most relevant to ad buyers: "When the platform that sells you the inventory also runs the agent that buys it, the agent works for the platform." According to Palmer, "Most companies will eventually need a strategy for all three."

Palmer names no company in that sentence. The description fits any platform that both owns advertising inventory and supplies the automated tools that buy it, as the largest advertising businesses do.

The org chart problem

The second half of the essay moves from procurement to management. "Almost every org chart in use today was designed for an all-human workforce," Palmer wrote. "The boxes are people. The lines are reporting relationships." Agents, the essay argues, break those assumptions: "They scale instantly, work continuously, cost compute instead of headcount, and treat supervision as a configuration setting."

According to Palmer, most companies are improvising the first step: "hybrid human/AI coworker teams, an agent embedded in a marketing pod with a person reviewing its output, and nobody quite sure whether the agent reports to the marketer, to marketing ops, or to the vendor."

The essay then sketches what comes next, an organisation with four kinds of workers on one chart:

  • Humans
  • Hybrid teams of humans and agents
  • Orchestrator agents, supervised by humans, that manage teams of other agents
  • Fully autonomous agents "that rarely touch a human at all, transacting with other companies' agents the way AAMP anticipates"

Each tier, Palmer argued, needs things the current chart has no box for: "who the agent reports to, who owns its output, who can shut it off, how its work gets reviewed, which budget line it lives on, and who answers for what it does." The essay points to information technology departments as a model. They "already treat software agents as non-human identities, each with a named human owner, scoped credentials, and a kill switch. Marketing needs the same discipline, on a chart everyone can read."

The four questions return here as what Palmer calls "org-design primitives." "Who decides the steps is a job description. Who approves the actions is a reporting line. Does it learn is a performance review. Who the agent serves determines whether it belongs on your chart at all."

Most large companies are some way from that picture. A WSJ Intelligence and Code and Theory study of 801 executives, publicised in October, found that one in four large US companies has no one accountable for coordinating AI, and that only 18% subject decision-making AI agents to accountability and reporting structures equivalent to those applied to human leaders. The report also called for a named human accountable for each agent making significant decisions, the requirement Palmer attributes to IT departments.

The essay's close is a recommendation rather than a finding. "Agentic marketing, no matter how you choose to define it, will require a redesign of your organization based on new first principles," Palmer wrote, ending with: "Each of these choices now requires a box on your reworked org chart. I'd start today." The Palmer Group, according to the same biography, advises Fortune 500 companies on AI strategy, implementation and governance, the kind of work the essay says companies now need.

Katsur's response

Katsur's post on November 3 quoted the essay and went one step further. "Shelly cites Gartner's 'agent washing': rebranding chatbots, copilots, and automation as agents. We also see APIs being relabeled as agents. A new name does not create agency," Katsur wrote.

The post restated Palmer's four questions in compressed form, with a small change to the last: "Who decides? Who approves? Does it learn? Whose interests does it serve?" It then set out the Tech Lab's own position. "At IAB Tech Lab, we encourage agentic workflows with purpose: What are we agentifying? Why? How? Do we share a definition of agentic? How will we measure success?" The post ended: "Plenty to be excited about. Let's pair that ambition with clarity and accountability."

Katsur named no company whose application programming interfaces were being relabeled, and the post does not mention AAMP by name, although the essay does. The caution is consistent with Katsur's earlier comments. On the AdTechGod podcast in December 2025, Katsur described agentic AI discussion as having reached a fevered pitch and said protocols could not fix misaligned incentives or bad actors.

The remark carries weight because IAB Tech Lab writes standards much of programmatic advertising runs on: when its head says a new name does not create agency, the point applies to companies building on its protocols too.

One reply under Palmer's November 2 post went further than Katsur. An account named SHaiLLEY, with the handle @shails, wrote: "Well said. The four questions are the basis around which AAMP is being designed and developed." The reply does not say who is behind the account or point to any AAMP document. Since IAB Tech Lab named AAMP in February 2026, eight months before the essay appeared, the claim can describe overlap between the questions and AAMP's design principles, not influence of the essay on that design.

What the essay does not establish

The essay applies no product to its own test. It quotes vendor definitions, not vendor products, so it does not say whether Agentforce Marketing, Agent Orchestrator, HubSpot's agents or Amazon Ads Agent would count as agents or automation under the first question, or which approval model each ships with. Nor does it accuse any named vendor of agent washing: the Gartner term is applied to "sales organizations" in general, and the seven companies appear as examples of conflicting definitions, not of rebranding.

Its market figures are dated, and the costs it describes go unmeasured. "Confusion at this scale comes at a price," Palmer wrote, without an estimate of wasted spend, failed deployments or compliance incidents. The four tiers of workers, finally, are a projection - "just over the next threshold," in Palmer's words - rather than a description of organisations already operating that way.

Why this matters for marketers

For advertisers, the essay puts the pressure on approval. Whether a system waits for human sign-off or ships on its own decides who carries the risk when an agent spends budget badly, misstates a product or breaks a platform policy, and Palmer's point is that two vendors can use the same word for both.

The timing matters too. The essay is aimed at 2027 planning, after Amazon renamed its ad platform after its agent and while IAB Tech Lab collects comments on AAMP 3.0. Katsur's questions - "Do we share a definition of agentic? How will we measure success?" - are posed as open ones by the body that writes the specifications.

For publishers, the fourth question has a specific edge. AAMP and AdCP are built for agents on both sides of a transaction, so Palmer's warning about platforms that sell inventory and run the buying agent bears directly on how those agents will be owned and audited. The definitions are being written before the spending arrives.

Timeline

Summary

Who: Shelly Palmer, Professor of Advanced Media in Residence at Syracuse University's S.I. Newhouse School of Public Communications and chief executive of the consultancy The Palmer Group, wrote the essay. Anthony Katsur, chief executive of IAB Tech Lab, endorsed it on X. The vendors whose definitions the essay compares are Salesforce, Adobe, HubSpot, Amazon, Bloomreach, Klaviyo and Profound.

What: An opinion essay, "What Agentic Marketing Actually Means," arguing that vendors use "agentic marketing" for incompatible things and disagree on whether a human approves an agent's work before it ships. It repeats Gartner's June 2025 "agent washing" estimate that roughly 130 of thousands of vendors offer genuine agentic capabilities, and its forecast that more than 40% of agentic AI projects will be scrapped by the end of 2027. It proposes four questions - who decides the steps, who approves the actions, whether the system learns, and whom the agent works for - and argues that companies will need org charts covering humans, hybrid teams, orchestrator agents and fully autonomous agents. Katsur added that APIs are also being relabeled as agents.

When: The essay is dated November 1, 2026. Palmer promoted it on X on November 2, and Katsur's post followed on November 3, a week before IAB Tech Lab's AAMP 3.0 comment period closes on November 10.

Where: On Palmer's website, shellypalmer.com, and on X. The essay addresses marketing technology buyers generally, with specific reference to IAB Tech Lab's AAMP protocols for advertising transactions between agents.

Why: Marketing and advertising vendors are selling products as AI agents while defining the term in ways that carry different risk, compliance and staffing consequences, especially on human approval. The essay matters because it gives buyers a short test for those claims during 2027 planning, and because the head of the standards body building the industry's agent protocols has publicly backed its warning that a new name does not create agency. The essay itself offers no product tests and no data newer than Gartner's 2025 figures.