This page is about the category, not the seat and not the engagement. If the question underneath is what a GTM engineer is and what the role now covers, that ground belongs to the explainer on marketing becoming an engineering function. This one assumes somebody has sent you a deck with the word agentic on it and you want to know whether it means anything.
exist, against thousands of vendors claiming them
will be cancelled by the end of 2027
improved their productivity, by 2028
The line already exists, and marketing did not draw it
The most useful definition of an agent in circulation was not written for go-to-market at all. It comes from Anthropic's engineering guidance on building agents, published in December 2024, and its value here is that it was written by people with no category to sell. It separates two things that get called by the same name:
"Workflows are systems where LLMs and tools are orchestrated through predefined code paths. Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks."Anthropic, Building effective agents, 19 December 2024
Read it twice, because the whole distinction sits in four words: predefined code paths. The question is not how much intelligence is in the system. It is when the sequence of steps was chosen, and by what. If an engineer decided the order before the run started, the system is a workflow, and it stays a workflow no matter how sophisticated the thing filling in each step happens to be.
This matters commercially because it is the one test a vendor cannot argue with. It is not about capability, quality or outcome. It is a structural fact about the software, and the person who built it knows the answer immediately.
What that makes agentic GTM
Apply the line to revenue work and most of the category sorts itself in about a minute.
A form fill triggers an enrichment, the enrichment feeds a score, the score picks a routing rule, the routing rule assigns an owner and starts a sequence. Every branch in that tree was drawn by a person and can be drawn again on a whiteboard. Put a model in the enrichment step to classify the account, and another in the copy step to write the first line, and the structure has not changed at all. That is automation with model inference in it, which is a genuinely useful thing and is not an agent.
The agentic version is different in kind. It is given an outcome rather than a path: qualified meetings in this segment this quarter. It decides which accounts are worth touching, which channel to use, what to do when a reply arrives that nobody anticipated, and when to stop spending attention on an account that is not going to convert. Anthropic is specific about when that shape is warranted:
"Agents can be used for open-ended problems where it's difficult or impossible to predict the required number of steps, and where you can't hardcode a fixed path."Anthropic, Building effective agents, 19 December 2024
Vendors inside the category define it in compatible terms when they are being careful. Tapistro describes agentic GTM as a system that "makes decisions, responds to live buyer signals, and coordinates action across channels without requiring human input at every step". That is a reasonable definition. The difficulty is that it describes an ambition, and nearly every product in the category can be made to sound like it in a deck.
Six tests you can run on the claim
These are questions about structure, not features. A feature question invites a demo. A structural question has one answer and the engineering team already knows it.
| Ask this | An agent answers | Rebranded automation answers |
|---|---|---|
| Who chose the order of the steps, and when? | The model, at runtime, differently on different accounts | An engineer, before the run, identically every time |
| Can you list every action it might take? | No, that is the point of it | Yes, they are in a configuration file |
| Can it decide to stop? | Yes, stopping is one of the decisions | No, it proceeds or it errors |
| What happens when conditions change mid-run? | It re-plans from the new state | It continues, or it breaks and waits |
| When it acts wrongly, what shows you why? | A record of the reasoning behind that choice | An activity log showing that the step ran |
| Does the demo show a decision or a sequence? | A judgement call you could disagree with | A sequence running faster than a human would |
The sixth is the cheapest and it works on a recording. Watch what the demo is proud of. If the impressive part is that many things happened quickly, you are watching automation, and automation running quickly is a good thing to buy. If the impressive part is that the system did something the presenter has to justify, you are watching an agent.
Checklists for this already exist and most are published by vendors. They tend to ask about capabilities rather than structure, which is a reasonable thing for a vendor to do and a weak thing for a buyer to rely on, because nobody publishes a test their own product fails.
The seventh question, which disqualifies most of the category
Everything above assumes the goal is to find a real agent. For most revenue work, that is the wrong goal.
Gartner's position is that many use cases positioned as agentic today do not require agentic implementations. It attaches no percentage to that, and neither will this page. Anthropic reaches the same conclusion from the engineering side, recommending "finding the simplest solution possible, and only increasing complexity when needed", and noting that workflows "offer predictability and consistency for well-defined tasks".
Most go-to-market plumbing is well-defined. Deduplicating records, routing a lead to the right owner, firing a sequence when a condition is met, stopping enrichment on rows that will never qualify. These jobs want software that does the same thing every time and costs the same every time. Making them agentic buys variance that somebody then has to supervise, at a price per run that a fixed path does not charge.
So the seventh question is the one to ask first: what would a well-built fixed workflow fail to do here? If there is a specific answer, naming a decision that cannot be written down in advance, the agentic version is worth pricing. If the answer is that it would be less impressive, the workflow is the correct build and the cheaper one.
Why the word got stretched
Gartner named the mechanism in June 2025. It calls the practice agent washing: rebranding existing products, including AI assistants, robotic process automation and chatbots, without substantial agentic capability. In the same release it put a number on the gap, estimating that roughly 130 real agentic AI products existed against the thousands of vendors claiming to offer them.
"Most agentic AI projects right now are early stage experiments or proof of concepts that are mostly driven by hype and are often misapplied."Anushree Verma, Senior Director Analyst at Gartner, 25 June 2025
Gartner's forecast from the same release is that over 40% of agentic AI projects will be cancelled by the end of 2027, attributed to escalating costs, unclear business value and inadequate risk controls. In November 2025 it published a second forecast aimed squarely at revenue teams: by 2028, AI agents will outnumber sellers tenfold, while fewer than 40% of sellers will report that agents improved their productivity.
Hold those two side by side and the shape of the next two years is legible. Deployment is forecast to run well ahead of reported benefit. Both are forecasts rather than measurements, which is worth saying because they are quoted everywhere as though they were findings. Nobody has published a controlled comparison of agentic go-to-market against a matched non-agentic baseline, and until somebody does, the honest position is that the category's effectiveness is unmeasured rather than proven or disproven.
What this page is deliberately not answering
Two adjacent questions come up immediately and both belong somewhere else.
- How you would actually run a go-to-market function if parts of it were agentic. That is an operating model question, it is the one most of the competing pages answer, and it is a different piece of work from defining the category. The opinionated version of that answer lives in GTM OS, which is a built system rather than an essay about one.
- Who does this work and what the role is called. The definition, the history and the hiring bar for the seat sit in the GTM engineer explainer, and the day-to-day version of the job is in what a Clay GTM engineer actually does all day.
The reason for drawing that boundary in the open is that the two get conflated constantly, and the conflation is what lets a vendor answer a definitional question with a product.
The practical versionIf what you actually have is a fixed path that nobody maintains, the agentic question is premature and the cheaper fix is upstream. The build-versus-buy decision underneath it is worked through separately.
Build vs buy GTM softwareQuestions people ask
What is agentic GTM?
Agentic GTM is go-to-market work where software decides what to do next rather than executing a path somebody wrote in advance. The useful definition comes from outside go-to-market, from Anthropic's engineering guidance on building agents, which separates two things people call by the same name: workflows are systems where LLMs and tools are orchestrated through predefined code paths, while agents are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks. Applied to revenue work, a sequence that fires on a trigger and then runs fixed steps is a workflow, however much model inference happens inside those steps. A system given an outcome, which then chooses the accounts, the channel, the timing and the moment to stop, is an agent. Most of what is currently sold as agentic GTM falls on the workflow side of that line.
What is the difference between agentic GTM and marketing automation?
Who chose the steps, and when. In marketing automation the path exists before the run starts: a form fill triggers an enrichment, a score, a routing rule and a sequence, and every branch in that tree was written by a person and can be drawn on a whiteboard. Adding a language model to the enrichment step or the copy step does not change the structure, because the model is filling in a cell inside a path it did not choose. In an agentic system the path is selected at runtime by the model. Anthropic's guidance is explicit about when that is warranted, saying agents can be used for open-ended problems where it's difficult or impossible to predict the required number of steps, and where you can't hardcode a fixed path. If you can hardcode the path, the honest label is automation, and automation is not a lesser thing.
How can you tell if a vendor's AI agent is really an agent?
Ask six structural questions rather than reading the feature list. Who chose the sequence of steps, the model at runtime or an engineer beforehand. Can the possible actions be enumerated in advance, because a finite list in a configuration file is a workflow. Can the system decide to stop, since ending is a decision and a pipeline can only proceed or fail. What happens when conditions change halfway through a run, re-planning or carrying on regardless. When it does something wrong, who is accountable and what record shows why it chose that action. And does the demo show a decision being made or a sequence being executed, which is the fastest of the six because a recorded demo cannot hide it. Vendor checklists for this exist, and they tend to ask about capabilities the vendor happens to have.
Does agentic GTM actually work?
The evidence available is about deployment rather than results, and the two forecasts that exist point in different directions. Gartner predicted in June 2025 that over 40% of agentic AI projects would be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls, and its analyst Anushree Verma described most current projects as early stage experiments or proof of concepts that are mostly driven by hype and are often misapplied. In November 2025 Gartner separately predicted that by 2028 AI agents would outnumber sellers tenfold while fewer than 40% of sellers would report that agents improved their productivity. Both are forecasts rather than measurements, and neither tells you whether a particular system works. Nobody has published a controlled measurement of agentic go-to-market against a matched non-agentic baseline, and that absence is worth stating plainly rather than filling with a case study.
Should every GTM workflow be agentic?
No, and this is the question most of the category avoids. Gartner states that many use cases positioned as agentic today do not require agentic implementations, without attaching a percentage to that claim. Anthropic's guidance says the same thing from the engineering side, recommending finding the simplest solution possible and only increasing complexity when needed, and noting that workflows offer predictability and consistency for well-defined tasks. Most revenue plumbing is well-defined: deduplicating records, routing a lead to an owner, firing a sequence when a condition is met. Those jobs want a system that does the same thing every time and costs the same every time. Making them agentic buys variance you have to supervise, at a price per run that a fixed path does not charge.
Sources
- The workflow and agent definitions, the guidance on open-ended problems and unpredictable step counts, and the recommendation to find the simplest solution possible · Anthropic, Building effective agents, 19 December 2024, read 24 September 2026
- Agent washing defined as rebranding AI assistants, RPA and chatbots without substantial agentic capability; roughly 130 real agentic AI products against thousands of vendors claiming them; the forecast that over 40% of agentic AI projects will be cancelled by end of 2027; and the Anushree Verma quotation · Gartner press release, 25 June 2025, quotations read via IT Pro's reproduction, 27 June 2025, both read 24 September 2026. gartner.com returns HTTP 403 to automated requests, so the wording here is taken from that reproduction and the press-release title
- The forecast that by 2028 AI agents will outnumber sellers tenfold while fewer than 40% of sellers report improved productivity · Gartner press release, 18 November 2025, read 24 September 2026
- A vendor definition of agentic GTM, quoted as an example of how the category describes itself · Tapistro, read 24 September 2026
Before the deck arrives
Want the fixed path built properly first?
Most teams asking the agentic question have a workflow nobody maintains and a bill nobody can explain. That is the cheaper problem and it is upstream of this one. Tell me which system matters most and you get example profiles and a call with your pick before anything starts.
Placement runs through AYTalent. Engineers are employed and trained on the bench before they start with you.