Most of what gets written about agentic marketing is either a vendor pitch wearing a thought-leadership hat or a warning written by someone who has never deployed an agent. I have been building these systems inside my own agency and my own client work for long enough to have opinions I can defend, and long enough to have been wrong in public a few times.
This is where I actually stand. Ten positions, each one earned rather than borrowed. Where a position needs proof, the proof is linked. Where I have already written the long version, that is linked too.
01The word “agent” is worthless as a buying signal
Gartner defines agent washing as labeling basic AI features like chatbots and assistants as agents, purely to create hype and drive sales interest. The software underneath does not change. The pricing page does.
This is not a minor annoyance. Gartner’s own estimate is that only around 130 vendors in the entire market deliver genuine agentic capability, against thousands claiming it. When a category is that polluted, the label tells you nothing and you have to test behavior instead.
Yoav Shoham, Stanford professor emeritus and AI21 Labs cofounder, framed the gap well in MIT Technology Review: the industry needs clearer expectations about what these systems do, how autonomously they operate, and how reliably they perform. Nobody is going to volunteer that clarity. You have to demand it.
// THE_LONG_VERSION: Agentic AI in Marketing: What’s Real, What’s Agent-Washing, and How to Stay in Control
02Automation is not the lesser thing
The correction to agent washing is not to sneer at automation. Automation is excellent, most marketing teams should be running more of it, and for predictable high-stakes tasks a deterministic workflow is often the safer and better choice.
The mistake is paying the autonomy premium for something that follows a fixed path you defined. Buy the workflow, price it like a workflow, and deploy it where its predictability is an asset rather than a limitation.
// THE_LONG_VERSION: Marketing Automation vs. Agentic Marketing: The Critical Difference
03The job changed from instruction to delegation
Think with Google’s position on agentic marketing is the most useful reframe I have read on this. Their argument is that prompt engineering is a tactic, not a strategy, and that the real shift is from instructing tools to delegating to agents. They describe the era of the tool operator as ending, replaced by the age of agent managers.
That is an operational change, not a vocabulary change. Getting better at prompting does not prepare you to manage a system that runs without you watching it.
04Govern an agent with the rigor you would give a new hire
The same Google analysis makes the point I keep coming back to. Because the autonomy is real, the governance has to be real: role-based access, explicit boundaries on what the agent may decide alone, and controls that cap operational and financial risk.
The instruction to resist anthropomorphizing these systems while still managing them like new hires is the correct tension to hold. The agent is not a person. The risk it carries is comparable to one.
// THE_LONG_VERSION: The Future of the CMO: Orchestrating Human and Machine Intelligence
05Stage, do not publish
The single most useful operating rule I have found comes from Involve Digital’s read on agentic workflows in 2026. Their framing is human-on-the-loop rather than human-in-the-loop: the human sets the parameters and reviews the exceptions, and the agent runs continuously inside those bounds instead of stopping for permission at every step.
Applied to content, their stated best practice is agents that generate and stage, with human approval required before anything publishes. The agent removes the creation work. The human keeps the decision about what goes live.
That distinction between staging and publishing is close to the only line I fully trust right now.
It separates people using autonomy to move faster from people who have handed away judgment they were not ready to give up.
// THE_LONG_VERSION: Agentic Content Strategy: Building Systems, Not Just Assets
06The failure that will hurt you is quiet
Research from Marrina Decisions on agentic marketing systems contains the stat that should reset everyone’s confidence: only 11 percent of organizations have implemented actual governance frameworks for their AI agents, despite deployment growing quickly.
Their description of the real risk is precise, and it is not the one people brace for. It is a well-intentioned agent taking a subtly incorrect action, at scale, repeatedly, with nobody noticing for weeks. It surfaces as drift in pipeline quality or win rate, not as a visible failure anyone catches in a dashboard glance.
Immutable audit logs, regular human review of decision patterns, and anomaly detection on your performance metrics are the defenses. All three are boring and all three have to exist before you scale, not after something breaks.
// THE_LONG_VERSION: Measuring the Unmeasurable: Analytics in the Age of AI Campaigns
07Adoption is not proficiency
McKinsey’s 2026 organizational research, summarized here, found 88 percent of organizations experimenting with agentic AI and 81 percent of those reporting no meaningful bottom-line gain. Both numbers are true simultaneously.
Gartner’s prediction that over 40 percent of agentic AI projects will be canceled by 2027 attributes the failures to unrealistic expectations and poor implementation rather than to the technology.
I read that as an argument for specificity, not retreat. Pick one bounded problem, instrument it against pipeline instead of hours saved, and be honest when the number does not move.
08Generative and agentic are not a choice you make
The framing that treats these as competing options is a category error. Generative models are a component inside agentic systems. Asking which one to adopt is like asking whether to use engines or cars.
// THE_LONG_VERSION: Agentic AI vs Generative AI: What’s the Difference (and Why It’s Not Either/Or) · What Is an AI Agent? Anthropic, OpenAI, and LangChain, Compared
09Speed multiplies whatever quality you feed it
An agent that can produce a thousand personalized assets will produce a thousand slightly wrong personalized assets just as happily. Autonomy does not improve your strategy, your positioning, or your judgment about what deserves to exist. It removes the friction that used to hide their absence.
This is why I treat these tools as sparring partners on the thinking and operators on the execution, never the reverse.
If your strategy was thin before you introduced agents, agents will publish that thinness faster and at greater volume. The upstream work did not get less important. It got more exposed.
// THE_LONG_VERSION: Using LLMs as Creative Collaborators, Not Just Typists · Marketing is Being Rewritten by Agents
10Experiment in the open, ship only what is proven
This is the position the other nine rest on. I run these experiments publicly, on my own site and my own accounts, and I write up the results whether or not they flatter me. The failed test is usually more useful to another marketer than the polished win, because the failure tells you where the edge actually is.
My clients do not get experiments. They get the things that survived them.
I would rather move slowly and be right than move fast and be full of air.
The short version
Test the behavior, not the label. Price a workflow like a workflow. Delegate instead of instructing, and govern what you delegate. Let the agent stage and keep the publish decision. Build the audit trail before you scale, because the failure that gets you will be quiet. Measure revenue, not hours saved.
And if you cannot state in one sentence what your agent decides on its own versus what it was told to do, you do not have an agent yet. You have a workflow with a new coat of paint.
Sources
- What is Agent Washing? — How Marketing Technology Works, citing Gartner’s definition
- Agent-Washing: How to Spot Hype and Separate Buzzwords from Real Agentic AI — PROS, including the Yoav Shoham quote originally published in MIT Technology Review
- Agent Washing: How to Tell a Real AI Agent From a Glorified Workflow — PM Northstar
- Agentic marketing: Managing AI agents, not prompts — Think with Google
- Agentic AI for Business Workflows 2026 — Involve Digital
- Agentic AI Systems: How to Design Marketing Workflows That Execute Themselves — Marrina Decisions
- Agentic AI’s Two-Front Problem — How Marketing Technology Works, citing McKinsey, The State of Organizations 2026
- Agent Washing: The Definition and a Buyer’s Scorecard — Digital Applied, citing Gartner’s April 2026 Hype Cycle