Executive Summary

Key research findings


Ask a marketer how they use agents in their workflows today, and you’ll get a measured response. They’re careful to keep humans at the controls for consequential decisions, while relying on agents to speed up execution and scale analytical work. In fact, 64% of marketers trust a human teammate over an agent for judgment calls, and 41% have already overridden or shut off an agent mid-task, most often fixing it by adding the context the agent was missing.

So far, agents pay off in capacity. They’ve freed marketers to redirect effort toward planning, positioning, audiences, and campaign design; 30% named this as their largest skill shift after deploying agents. Today’s marketing teams are settling into a predictable division of labor: agents support repeatable execution, while people make source inputs, review routes, escalation, and rollback visible enough to stay accountable for what reaches customers.

Interesting differences arise when comparing mid-market businesses to enterprises. Mid-market teams tend to be more reactive to how their agents perform and drive value. They’re more doubtful that agents can run campaigns end to end without supervision (28% vs. 13%), they cycle through agents faster, and are much more likely to have downsized their teams or slowed hiring after deploying agents.

Enterprise teams, on the other hand, take a more conservative approach: they’re far less likely to rely on standalone LLMs (13% vs. 41%), are less likely to have made HR decisions around agents, and are more likely to have redirected people toward strategy when agents take on rote work.

This report explores how teams deploy AI agents for marketing across today’s lifecycle space, including how they use, monitor, and govern agents and whether they’ve bought into messaging around agentic AI in marketing and an autonomous, agent-led future. It sits alongside G2’s broader sales and marketing AI research.

Chapter 01

How autonomous are AI marketing agents in 2026?

AI agents for marketing teams aren’t running lifecycle programs end to end. Three in five marketers (61%) say agentic marketing only partly lives up to the hype: it’s great at bounded tasks, but it still needs a lot of oversight. And when a call depends on context, nuance, or accountability, 64% would rather trust a teammate.

Finding 1.1

Vendor claims versus reality

A perceived need for human oversight means agents only partially live up to marketing claims

61%

say agents only partly live up to the claims: they’re useful for bounded tasks but need heavy oversight

Marketers want faster execution, but not at the cost of control over what customers see. Three in five (61%) say agents only partly live up to the claims made for them: they help with bounded tasks, but teams still keep a close eye on them instead of handing off whole workflows. Only 7% say agents fall short outright. Company size doesn’t change the picture much: 65% of mid-market and 62% of enterprise marketers say agents only partly deliver.

Marketers see real value in speeding up well-defined work, and they want a person reviewing anything that shapes a marketing decision. For vendors and marketing leaders, that’s a clear signal: transparent, controllable help beats promises of hands-off, end-to-end autonomy. Regulators have drawn the same line: the FTC’s Operation AI Comply targets companies that overstate what AI can do.

  1. 01 30%

    Human judgment remains the boundary

    The claim marketers doubt most (30%) is that agents can replace human marketers, judgment, and emotional intuition. Their concerns go beyond the quality of the output.

  2. 02 34%

    Brand handoff depends on explicit inputs

    About a third (34%) say agent drafts capture brand tone well when teams give them detailed guidelines and examples, which makes good inputs a practical condition for wider use.

  3. 03 52%

    Mid-market teams do more of the editing

    More than half (52%) of mid-market marketers say agent drafts need human editing to sound on-brand, compared with 36% at enterprises.

For the most part, [agents] live up to the claims, but completely eliminating a human review from an expert would not be wise and not something that I would fully trust at this stage. There are certain lower level functions that are really just sharing information that I would be more likely to trust AI with. then generating language or content that needs to be strictly regulated.
Marketer Assessments of AI Agents Against Marketing Claims n = 100
Partially; assists with bounded tasks but requires heavy oversight
61%
Yes, lives up to claims for speed and productivity
23%
No, falls short of autonomous end-to-end marketing claims
7%
Other
9%
Depends on what kind of claims we're talking about. If you hear the ones the hyped ones that agents can do everything from A to Z, I don't think we see that. I think products that have to interact that have to do that bridge between the provider, the vendor and the customer still need a human in the loop.

Finding 1.2

AI Agents Living Up to Claims by Marketing Role

Across lifecycle, operations, brand, and product and growth roles, 60% to 82% see agents as useful but oversight-heavy. The mixed Other group is the exception. Build review gates that fit each function instead of assuming autonomy carries over from one team to the next.

AI Agents Living Up to Claims by Marketing Role n = 100
Yes, lives up to claims for speed and productivity Partially; assists with bounded tasks but requires heavy oversight No, falls short of autonomous end-to-end marketing claims Other
View data table
AI Agents Living Up to Claims by Marketing Role
Category Yes, lives up to claims for speed and productivity Partially; assists with bounded tasks but requires heavy oversight No, falls short of autonomous end-to-end marketing claims Other
Lifecycle, CRM, and email marketing (n=15) 4 9 2 0
Marketing leadership, generalist, or multi-channel operations (n=44) 11 28 3 2
Brand, content, and communications strategy (n=16) 4 10 0 2
Product, growth, and acquisition marketing (n=11) 2 9 0 0
Other (n=14) 2 5 2 5
Respondent counts, n = 100

Finding 1.3

Exploring the limits of trust in AI agents

When it comes to judgment calls, humans still win over agents

64%

trust a human teammate’s judgment over an agent’s, pointing to context, nuance, and accountability

Marketers want speed, but they aren’t ready to hand over the decisions that shape customer relationships. Nearly two-thirds (64%) trust a teammate more than an agent for judgment calls, pointing to the context, nuance, and accountability people bring when a choice affects the brand or the customer. Company size makes little difference: 61% of mid-market and 64% of enterprise marketers say the same.

Some trust comes with conditions: 24% decide task by task, and 5% favor agents for data processing, reporting, and math. Judgment stays in human hands. The practical move is to deploy agents where people can check the rules and results, and keep a clear human owner for decisions that depend on reading a customer’s situation.

  1. 01 56%

    Trust is earned task by task

    More than half (56%) say the agent they trust most has since taken on broader or more complex work, so a tightly scoped start doesn’t have to stay that way.

  2. 02 3%

    Few teams pull back

    Only 3% cut back or tightly limited an agent’s scope after it made mistakes. Once an agent earns its place, teams tend to keep giving it more.

If it was something that was more of a data driven call, I would say I'd definitely trust the agent. Now if it was the nuances within the business, then understanding, you know, maybe the brand or our actual customers and growers, I would have to say that I would definitely trust my team's judgment over an agent.
Trust in AI-Agent and Human Judgment Calls n = 98
Trusts human teammates more due to context, nuance, and accountabilityTrusts humans and agents equally or conditionally depending on the taskTrusts the agent more for data processing, reporting, and mathOther
When it comes to the analytics reporting agent specifically, I trust it more than people on the team because the agent is significantly better at data analysis, mathematics, and pattern recognition than any person could. In fact, even when I was doing it, I could only look at one section of the data at a time, whereas the AI's agent is able to look at the data set as a whole even though it goes over up to 2 years.

Finding 1.4

AI Agent Versus Human Judgment Trust by Marketing Role

Judgment stays human-led in every role group, but conditional trust varies: 40% of product and growth marketers trust people or agents depending on the task, compared with 7% of lifecycle marketers. These groups are small, so treat this as directional. Delegation rules should follow the workflow, not just the job title.

AI Agent Versus Human Judgment Trust by Marketing Role n = 98
Lifecycle, CRM, and email marketing (n=15)Marketing leadership, generalist, or multi-channel operations (n=44)Brand, content, and communications strategy (n=15)Product, growth, and acquisition marketing (n=10)Other (n=14)
Trusts the agent more for data processing and math22100
Trusts humans more for context, nuance, and accountability1129869
Trusts humans and agents equally, sometimes depending on the task112641
Other11004
Smaller share of segmentLarger share
View data table
AI Agent Versus Human Judgment Trust by Marketing Role
Category Trusts the agent more for data processing and math Trusts humans more for context, nuance, and accountability Trusts humans and agents equally, sometimes depending on the task Other
Lifecycle, CRM, and email marketing (n=15) 2 11 1 1
Marketing leadership, generalist, or multi-channel operations (n=44) 2 29 12 1
Brand, content, and communications strategy (n=15) 1 8 6 0
Product, growth, and acquisition marketing (n=10) 0 6 4 0
Other (n=14) 0 9 1 4
Respondent counts, n = 98

TL;DR: 61% of marketers say AI agents only partially live up to marketing claims, and 64% trust a human teammate over an agent for judgment calls.

Chapter 02

How do marketers monitor and override AI agents?

Agents run under human supervision. Four in ten marketers (41%) have already stepped in to override or shut off an agent mid-task, and the most common fix was simple: give it the context it was missing and run it again. Marketers have also drawn firm lines around what agents can’t touch: regulated data, sensitive customer moments, pricing and offers, and brand voice.

Finding 2.1

The override moment

Four in ten marketers have already overridden or shut off an agent

41%

have overridden or shut off an agent while it was running

Marketers are used to stepping in as agents carry out their day-to-day work. Four in ten (41%) have overridden or shut off an agent mid-task. Most of those moments were routine course corrections: of the 40 who walked us through one, 38% adjusted the prompt, added context, or reran the task, while 23% shut the workflow down, reversed it, or switched the feature off.

Several of the most detailed stories follow the same arc. The agent acted on the data in front of it, missed customer context it couldn’t see, and a person caught the gap. Agents that work from fuller customer data, with pause and rollback built in, address the failure marketers describe most clearly.

  1. 01 16%

    Targeting and journey errors lead the list

    Asked about the worst thing an agent had done on its own, 16% named a targeting, segmentation, or journey error, and 15% pointed to off-brand or inappropriate messaging.

  2. 02 18%

    Overrides leave stronger controls behind

    Among those who described an override, 18% responded by tightening guardrails, rules, and QA checkpoints.

One of the agents we were running, it was over prioritizing a customer segment based on recent digital engagement, but it didn't have enough context from the broader customer history. So it started recommending follow up that felt a little too aggressive for the relationship and really just wasn't appropriate for a luxury customer journey, right? Luxury is meant to be low pressure, kind of environment. So we caught it through human review before anything material went out, and then we paused that workflow. We looked at the decision logic and inputs and really tightened the rules around recency, customer status and frequency of contact.
It was in regards to creative design work. So we have a design agent and that agent was just taking liberties and basically hallucinating backgrounds and not following the prompts that it was encoded with. So I had to shut it down in order to stop utilizing tokens that would have been wasted.

Finding 2.2

The override moment, by company size

Why enterprise teams rarely hit the override point

Nearly two-thirds (62%) of Enterprise teams have never overridden or shut off an agent, compared with 48% of mid-market teams. The difference can be chalked up to Enterprises’ risk aversion: larger companies tend to keep agents on a short leash, limiting them to narrow, low-stakes, or internal-facing tasks with human approval needed for anything that goes live to customers. Because of these guardrails, override moments rarely occur.

Similarly, mid-market teams who face fewer override moments say they devote more time and attention to hands-on tuning with their agents to prevent missteps: refining instruction files, editing prompts, or pausing workflows to make tweaks.

I think we’re very conservative when it comes to autonomous AI operations. We do not allow it. The leash is short. So we have more control and governance around what we are trying to test in utilizing AI for, as opposed to allowing it to be autonomous and driving our marketing capabilities.
We have not come close to stepping in because we’re really only comfortable deploying these agents [on] some of these lower risky tasks. We haven’t built up the confidence internally yet from a organizational perspective to deploy them for more widespread or for higher intensive tasks.
I don’t think that we’ve provided enough access necessarily for an agent to require any override protocols… I believe that as a safeguard, there is still kind of a four eye check where there’s a human approval prior to going live.

Finding 2.3

Decisions marketers keep off-limits

Marketers keep regulated data, sensitive customer moments, pricing, and brand voice away from agents

4%

say no category of decision is off-limits to agents

We asked marketers whether there are decisions an agent should never touch, even when it’s technically capable. Only 4% said nothing is off-limits; the vast majority do place restrictions on where agents should get involved. The lines they draw are consistent: 23% keep confidential, legal, regulatory, and customer data away from agents; 17% save complaints, churn, and other sensitive relationship moments for people; 16% hold back pricing, offers, and budget decisions; and 16% keep brand voice and external executive communications in human hands. The data category tracks the obligations marketers already carry under frameworks such as the General Data Protection Regulation (GDPR) and the Health Insurance Portability and Accountability Act (HIPAA).

These are the areas where one wrong call costs more than the time saved. Marketers are drawing these lines now, before agents get more capable, and platforms that make those boundaries easy to set and audit will be easier to trust.

  1. 01 27%

    Some agents don’t survive production

    More than a quarter (27%) have retired an agent because of poor performance, inaccurate output, or its limitations, so failed pilots sit right alongside the agents that scaled. Mid-market teams are more likely to have pulled the plug: 33%, vs. 19% at enterprises.

  2. 02 4%

    Almost no one gives agents free rein

    Only 4% said no category of decision is strictly off-limits for agents.

Yes. There are a few areas where I deliberately keep a human in control. So things like pricing and offers, customer complaints, churn decisions, really anything that could materially affect the customer relationship.
While [the agent] generated text quickly, it consistently struggled with the nuance tone required for sensitive patient journeys. Requiring extensive human rewriting. So we realized our team spend more time correcting clinical context and tone than drafting from scratch.
I think to do with pricing and dollar amounts, I'm just very cautious and even like our budget. All these things are still manually managed.

TL;DR: 41% of marketers have already overridden or shut off an agent mid-task, and most overrides end with stronger guardrails rather than abandonment; only 4% give agents free rein.

Chapter 03

What tasks do marketers actually delegate to AI agents?

Drafting is where agents earn their keep. Among the AI agent use cases marketers named, content generation is the top use, helping teams turn out campaign materials faster without giving up control of the campaign. Anything customer-facing still gets a human sign-off, and 39% require several rounds of review before it goes live.

Finding 3.1

The current state of AI agent adoption

Content generation is the leading agentic deployment in lifecycle marketing

26%

name content generation, copywriting, and creative drafting as their main autonomous use

Marketing teams want campaigns out the door faster, but most aren’t ready to hand over end-to-end control. Instead, they limit agents to working autonomously on specific tasks, most often content generation and creative drafting (26%), as well as segmentation and data workflows (17%).

Another 26% describe limited or no autonomous use, keeping agents as supervised assistants. Drafting makes a natural starting point because teams can review everything before it reaches customers. Start with agents where review stays practical, then expand as teams build confidence in the controls.

  1. 01 30%

    Governance has to travel beyond the platform

    Three in ten (30%) say standalone LLMs and agents are their main tool, so review, data-use, and approval practices need to reach beyond the core marketing stack. Mid-market teams lean on these tools most: 41%, vs. 13% at enterprises.

  2. 02 20%

    Internal reporting builds early trust

    For 20%, the first task they let an agent run with minimal oversight was reporting, data consolidation, or analytics: internal work where they could build confidence before anything customer-facing.

There are also certain iterative tasks such as you know, repetitive newsletters, monthly social posts rounding up different events we'll be at, things like that where I have an agent working on producing those on a monthly basis. And then I review and finalize.
We have tried to use AI agents to help automate and our workflow, starting from understanding the customer, orchestrating the customer journey, developing and the right content, and selecting the right channels for the different customers based on the data we have.

Finding 3.2

Autonomous AI Agent Deployment Tasks by Company Size

Content drafting is the top autonomous use case across business size (26% of mid-market and 28% of enterprise teams), and about a quarter of each group limits agents to working in supervised assistant roles. Beyond drafting, there’s a split: mid-market teams more often hand agents segmentation, lead scoring, and data workflows (24% vs. 13%), while enterprise teams more often give them QA, reporting, and admin work (19% vs. 7%).

Autonomous AI Agent Deployment Tasks by Company Size n = 93
Content generation, copywriting, and creative drafting Limited or no autonomous deployment; human-supervised assistance only Campaign planning, research, and competitive intelligence Quality assurance, reporting, and administrative tasks Segmentation, lead scoring, and data workflows Other
View data table
Autonomous AI Agent Deployment Tasks by Company Size
Category Content generation, copywriting, and creative drafting Limited or no autonomous deployment; human-supervised assistance only Campaign planning, research, and competitive intelligence Quality assurance, reporting, and administrative tasks Segmentation, lead scoring, and data workflows Other
Mid-market (500-4,999 employees) (n=46) 12 11 2 3 11 7
Enterprise (5,000+ employees) (n=47) 13 11 2 9 6 6
Respondent counts, n = 93

Finding 3.3

AI agent oversight models, failure modes, and guardrails

Agents typically work within strict guardrails

39%

require multi-stage human approval before anything goes out externally

Publishing creates a real tension: teams want agents to speed up content work, but someone still has to answer for what customers see. Well over a third (39%) of marketers require mandatory multi-stage human approval, the most common of the AI guardrails marketers described.

That puts the human checkpoint where the consequences are: the moment work reaches customers. It matters because a public mistake can outlive the workflow that made it. Teams should set clear review routes for anything public-facing and leave the prep work to agents.

  1. 01 32%

    Approval is one control among several

    Nearly a third (32%) monitor agents through performance metrics, A/B tests, and downstream KPIs, judging agent work by its results after launch as well as before.

  2. 02 26%

    Trust depends on validating inputs

    For 26%, the worst thing an agent had done was a factual error, a made-up metric, or mixed-up data, which is why data checks need to sit alongside approval.

The email creation is live in production, but it is then passed to a human being for review and with a legal and compliance lens and then put into the tool for deployment.
AI Agent Guardrails and Review Steps n = 99
39%Mandatory multi-stage human approval before external publishing
20%Explicit reference files, prompt rules, and brand guidelines
14%Data access restrictions, role permissions, and privacy boundaries
6%
20%Other
Mandatory multi-stage human approval before external publishing (39%)Explicit reference files, prompt rules, and brand guidelines (20%)Data access restrictions, role permissions, and privacy boundaries (14%)Post-generation automated checklists, QA agents, and error alerts (6%)Other (20%)
We have multiple levels of review, obviously, we when we work at a financial institution, there are multiple levels. So you have like proofreading, basic, and like from the creative team, you also have the strategist check it, you also have compliance, and you also have the business unit. So it's passing through multiple hands. Before it ever goes public.

Finding 3.4

Agent Guardrails and Review Steps by Company Size

Multi-stage human approval is the most common guardrail for both mid-market (33%) and enterprise teams (43%). The clearest difference is how teams steer agents before review: mid-market teams lean more on reference files, prompt rules, and brand guidelines (26% vs. 13%). That fits the pattern in Finding 2.2: mid-market teams more often hand agents a rulebook and let them work within it, while enterprise teams lean a little harder on people in the loop. It’s the same shorter leash that leaves fewer enterprise teams ever needing to override an agent.

Agent Guardrails and Review Steps by Company Size n = 92
Explicit reference files, prompt rules, and brand guidelines Mandatory multi-stage human approval before external publishing Data access restrictions, role permissions, and privacy boundaries Post-generation automated checklists, QA agents, and error alerts Other
View data table
Agent Guardrails and Review Steps by Company Size
Category Explicit reference files, prompt rules, and brand guidelines Mandatory multi-stage human approval before external publishing Data access restrictions, role permissions, and privacy boundaries Post-generation automated checklists, QA agents, and error alerts Other
Mid-market (500-4,999 employees) (n=46) 12 15 7 2 10
Enterprise (5,000+ employees) (n=46) 6 20 6 4 10
Respondent counts, n = 92

TL;DR: Content generation leads autonomous agent deployment at 26%, and 39% of marketers require multi-stage human approval before agent work goes out.

Chapter 04

What results do AI agents deliver in lifecycle marketing?

So far, the biggest payoff is speed. More than half (55%) say faster campaign launches and production are the clearest change, while proof of AI marketing ROI, like conversion lift, is still thin. With production moving faster, marketers are putting more of their time into strategy, positioning, audiences, and campaign design.

Finding 4.1

Measured outcomes and hidden tradeoffs of agent use

AI agents' greatest impact is accelerating campaign production

55%

say faster campaign launches and production turnaround are the clearest change since deploying agents

Marketing teams want to get from idea to live campaign quickly, but speed alone doesn’t prove customer outcomes improved. More than half (55%) say faster campaign launches and production turnaround are the clearest change since deploying agents, which puts agents’ most visible value in making and shipping campaigns. That holds at both company sizes: 58% of mid-market and 55% of enterprise marketers named faster launches.

Better downstream performance came up far less often: just 10% named conversion, click-through, or response rates as the clearest change. Agents are freeing up capacity, but their effect on marketing results still needs proving. Judge agents first on cycle time, then test downstream effects on purpose.

  1. 01 15%

    Speed can dilute the work

    A smaller group (15%) found that faster production came with generic, unoriginal, or off-brand output, so brand review still matters even where agents speed things up.

  2. 02 8%

    Short-term gains can carry customer costs

    A small share (8%) saw gains in volume or open rates come at the expense of tone, brand trust, retention, or unsubscribe rates, so teams need to check customer experience alongside whatever metric an agent is optimizing.

The clearest change has been in time to complete routine work and the volume we are able to process. So tasks like verification, duplicate detection, consistency checks, and initial content drafts, move much faster.
Reported Metric Changes Following AI Agent Deployment n = 99
55%
12%
10%
5%
18%
Faster campaign launch time and production turnaround (55%)No clear or measurable change in tracked metrics (12%)Improved conversion, click-through, or response rates (10%)Improved deliverability or lower unsubscribe rates (5%)Other (18%)
I also was able to increase my open rates and click through rates, in some of the emails, also I ran some of email design choices by an agent, and the agent helped me increase those especially click through rates by, you know, 2 or 3% in a pretty, you know, like a 20,000 person email list, which are significant for that amount of people.

Finding 4.2

Tracked Metric Changes After Agent Deployment, by Industry

Faster launches were the top reported change in every industry, though financial services respondents cited them less often than B2B SaaS respondents: 40% vs. 62%. These groups are small, so confirm cycle-time gains in your own team before setting expansion targets.

Tracked Metric Changes After Agent Deployment, by Industry n = 99
B2B SaaS & tech (n=26)Healthcare & pharma (n=17)Retail & CPG (n=11)Financial services (n=10)Industrial & mfg. (n=14)Other (n=21)
Faster campaign launch time and production turnaround16964910
Improved deliverability or lower unsubscribe rates310100
No clear or measurable change in tracked metrics221214
Improved conversion, click-through, or response rates132211
Other422136

1. B2B SaaS, enterprise software, and technology services (n=26); 2. Healthcare, pharmaceuticals, and medical devices (n=17); 3. Retail, e-commerce, and consumer packaged goods (n=11); 4. Financial services, banking, and insurance (n=10); 5. Industrial, manufacturing, energy, and business services (n=14); 6. Other (n=21)

Smaller share of segmentLarger share
View data table
Tracked Metric Changes After Agent Deployment, by Industry
Category Faster campaign launch time and production turnaround Improved deliverability or lower unsubscribe rates No clear or measurable change in tracked metrics Improved conversion, click-through, or response rates Other
B2B SaaS, enterprise software, and technology services (n=26) 16 3 2 1 4
Healthcare, pharmaceuticals, and medical devices (n=17) 9 1 2 3 2
Retail, e-commerce, and consumer packaged goods (n=11) 6 0 1 2 2
Financial services, banking, and insurance (n=10) 4 1 2 2 1
Industrial, manufacturing, energy, and business services (n=14) 9 0 1 1 3
Other (n=21) 10 0 4 1 6
Respondent counts, n = 99

Finding 4.3

How agents change marketer work and team capacity

After deploying agents, marketers shift efforts to strategic planning

30%

now spend more time on strategic planning, positioning, and campaign design

Marketers want the time agents save to go into better campaign decisions. Three in ten (30%) named strategic planning, positioning, and high-level campaign design as their biggest skill shift. Marketers describe moving away from assembling inputs and assets and toward decisions about audiences, messaging, and campaign design.

Agents don’t create that strategic time on their own: 20% now spend more time reviewing, fact-checking, and editing AI output. Some human attention moves up to strategy, and some goes into checking what agents produce. Plan for both shifts when sizing an agent program.

  1. 01 38%

    Agent value sits in information work

    Close to two in five (38%) say agents are best at data synthesis, research, and fast information retrieval, which makes a case for starting with work you can easily check.

  2. 02 48%

    Staffing plans mostly hold

    Nearly half (48%) report no change in headcount or hiring plans because of agents, while 20% report cuts or slowed hiring. Cuts are far more common at mid-market companies (30% vs. 9% at enterprises), even though more of those companies are growing overall (38% vs. 23%).

It's more about the proper documentation and data, right? Those are the skills that I'm training my team is to make sure that as structured data and as more structured workflows, the better the AI agent is going to be working. So it's less prone for editing and validation. So those are some of the skills we're working on.
Skills Receiving More Time After Agent Deployment n = 97
Strategic planning, positioning, and high-level marketing design
29
Reviewing, fact-checking, and editing AI outputs
19
Prompt engineering, directive writing, and agent building
12
Data governance, workflow architecture, and guardrail design
7
No change in skill allocation
3
Other
27
We are spending much more time on the strategic questions on how often should we be messaging our customers, which updates do they want to see? How can we improve the creative to make the email and web reading experience even better for our customers, whereas a lot of that time before was wasted in the information gathering stage of reconciling documents, compiling updates, getting approval from multiple stakeholders, having to do last minute updates as roadmaps changed.

Finding 4.4

Skills Receiving More Time After Agent Deployment, by Company Size

Enterprise teams are more likely to have shifted effort into strategic planning, positioning, and campaign design (38% vs. 22%). Mid-market answers spread more evenly across reviewing AI output, prompt work, and governance, and none of those smaller gaps reaches 10 points.

Skills Receiving More Time After Agent Deployment, by Company Size n = 91
Strategic planning, positioning, and high-level marketing design Reviewing, fact-checking, and editing AI outputs Prompt engineering, directive writing, and agent building No change in skill allocation Data governance, workflow architecture, and guardrail design Other
View data table
Skills Receiving More Time After Agent Deployment, by Company Size
Category Strategic planning, positioning, and high-level marketing design Reviewing, fact-checking, and editing AI outputs Prompt engineering, directive writing, and agent building No change in skill allocation Data governance, workflow architecture, and guardrail design Other
Mid-market (500-4,999 employees) (n=46) 10 8 7 2 5 14
Enterprise (5,000+ employees) (n=45) 17 10 5 1 2 10
Respondent counts, n = 91

TL;DR: 55% of marketers say faster campaign launches are the clearest change after deploying agents, while 8% have seen an agent improve one metric at the expense of tone, brand trust, or retention.

Chapter 05

Where will the agent-human boundary sit a year from now?

A year from now, 56% of marketers expect agents to handle routine execution while people keep the strategy and the final sign-off. Few expect AI agents to take over sensitive decisions within a year: just 4% say no category of decision is strictly off-limits to agents. The goal is to build workflows that make it safe to hand over more, while keeping sensitive customer moments in human hands.

Finding 5.1

The future boundary between agent execution and human judgment

Marketers expect a future of workhorse agents with humans retaining strategic approval

56%

expect agents to run routine execution while humans keep strategy and approval

Marketing teams want more capacity without giving up the decisions that shape customer experience. Looking a year ahead, 56% expect agents to handle routine execution and orchestration while people keep strategy and approval. Mid-market and enterprise teams are nearly identical here: 57% and 54%.

That’s a deliberately lopsided model: agents scale the recurring work, and people stay accountable for direction and sign-off. Design agents around clearly bounded workflows and approval points, and treat what an agent can do as separate from what it’s allowed to decide. A smaller group of enterprise marketers (15%) expects agents to take over analytics, segmentation, and QA while people keep creative and brand voice, a view no mid-market respondent shared.

  1. 01 29%

    Sensitive customer moments stay human

    Nearly three in ten (29%) named direct customer communication, complaint handling, and crisis response as work they haven’t handed to an agent.

  2. 02 15%

    Context unlocks the next handoff

    We asked what would have to change before they’d hand off a task they hold back today. Nearly three in ten (29%) want a proven track record of accuracy, and 15% name better context and data, like customer history, real-time data, or connected sources, as often as they name brand voice.

And so I see a future 12 months down the road where AI agents will handle a lot of the hard lifting work of identifying targeted customers via our CRM, making some informed updates to enablement based on some of the critical roadmap updates.
Humans, I think, will be more interested in orchestrating unique journeys or special one-off campaigns. Like always-on flows, I think could be more AI driven, whereas one-off campaigns where it's a special incentive for a limited time would probably need to be owned by a human who has a creative, innovative idea.

Finding 5.2

Expected Agent–Human Work Boundaries by Marketing Role

Preferences around agent guardrails may vary widely based on department. For example, product and growth marketers most often expect agents to execute while people approve (82%), while brand, content, and communications strategists would prefer more human approvals, particularly in external touchpoints. These signals are worth considering for any governance pilot that touches brand voice or customer interactions.

Expected Agent–Human Work Boundaries by Marketing Role n = 95
Agents handle routine execution and orchestration; humans retain strategy and approval The boundary will remain largely where it is today Humans retain customer-facing relationships, sensitive moments, and accountability Agents take on data analytics, segmentation, and QA; humans own creative and brand voice Other
View data table
Expected Agent–Human Work Boundaries by Marketing Role
Category Agents handle routine execution and orchestration; humans retain strategy and approval The boundary will remain largely where it is today Humans retain customer-facing relationships, sensitive moments, and accountability Agents take on data analytics, segmentation, and QA; humans own creative and brand voice Other
Marketing leadership, generalist, or multi-channel operations (n=43) 25 3 4 2 9
Lifecycle, CRM, and email marketing (n=12) 7 1 1 2 1
Brand, content, and communications strategy (n=15) 5 2 3 2 3
Product, growth, and acquisition marketing (n=11) 9 0 1 0 1
Other (n=14) 7 0 1 1 5
Respondent counts, n = 95

TL;DR: 56% of marketers expect humans to keep strategy and final approval a year from now, and just 4% say no category of decision is strictly off-limits to agents.

How the findings connect

Strategic Takeaways

01

Delegation grows when teams separate execution from judgment

A practical division of labor explains the gap between broad agentic marketing claims and what teams actually do in production. Teams delegate bounded drafting work, but retain human review where customer-facing consequences and judgment calls arise. The result is selective delegation shaped by accountability.

02

The realized value of agents is capacity

Right now, marketers most clearly link agents to faster campaign production, and they’re redirecting effort toward higher-level planning and design. Marketers’ expected future boundary preserves this asymmetry: agents scale routine work, while people continue to own strategy and approval.

Summary

Conclusion

The central change is a more deliberate division of labor: agents contribute to repeatable execution while people retain judgment at the moments that shape customer experience. The most credible path forward is governed delegation.

Challenges

The cost of getting customer-facing work wrong still keeps trust in check. Sensitive direct communication is a particularly firm boundary, with 29% naming complaint handling, crisis response, or comparable customer moments as work they haven’t handed to agents yet. Approval, factual validation, monitoring, and rollback need to operate together.

Forward looking

Teams can broaden agent responsibility by proving controls in bounded workflows first, then learning which tasks, inputs, and review routes make expansion reliable. Teams should measure success by the capacity agents create for stronger planning and customer-experience stewardship, alongside the quality of the work they release.

The durable agentic marketing model runs on human accountability.