Who Will Pick Your Eight Agents? Change Futurist | September 10, 2026

Identify the eight AI agents your marketing organization should run next year. Then name the person who picks them. Most teams can produce the first list in an afternoon and stall on the second question for a quarter (or more).

Four publications inside five days describe that stall from different angles. Joe Stanhope and Jessica Liu at Forrester report that 83% of B2C marketing decision makers now implement agentic AI in their workflows and 85% say it delivers meaningful business value, then state the constraint plainly: no marketer can run an infinite number of agents. Marketing Week's Secret Marketer describes a manager layer shaped by years of channel-first work and unable to lift a conversation from execution to strategy. Doug Chung and four McKinsey coauthors argue in HBR that agentic AI has erased the working boundary between marketing and sales while most companies keep deploying agents separately inside each function. And Charlotte Rogers reports that 71.2% of B2B CMOs believe their business values brand marketing while only 54.5% of marketing directors and vice presidents say the same.

Read those together and one finding sits underneath all four. Every AI decision now on a CMO's desk is a selection problem, and selection runs on judgment.

My August 31 edition argued that AI programs need a named owner. September 3 argued that the owner needs a measurement instrument. September 7 added the cost denominator and the routing rule that follows from it. This edition names the person who has to apply that rule. In most marketing organizations, that person manages a channel.

I have watched a Fortune 500 marketing team spend six weeks scoring 40 agent candidates on a weighted matrix, then hand the ranked list to a group of directors who debated the top three for an hour and approved all 40. Nobody in the room had ever had to defend a subtraction.

Cap the agent count before your vendors set it for you

Stanhope and Liu published AI Agents For Marketing Are Here: Are You Ready? on September 8. Their adoption numbers describe a market past the pilot stage. Eighty-three percent of B2C marketing decision makers actively implement agentic AI into their workflows. Eighty-five percent report meaningful business value. With Claude and ChatGPT in general use and most martech vendors shipping proprietary agents, access has stopped being the variable.

Their constraint is the useful part. Agent opportunities span customer insights, content creation, operations, and campaign execution, and each agent consumes resources to implement, maintain, use, and monitor. That makes the agent portfolio a finite budget line with a fixed ceiling, and somebody sets the ceiling.

Two forces will set it for you if you leave the question open. Vendors ship agents inside platforms you already own, so the count grows through renewals nobody reads as a portfolio decision. And every team that wants one can now build one, which converts agent proliferation into a shadow IT problem wearing a marketing badge.

That maps directly onto the mechanism axis in my AI capability framework. Use the lightest mechanism that clears the task's accuracy bar, and reach toward agentic execution only where the task's openness requires it. September 7 established the cost consequence of that rule. Forrester establishes the operational one. Each agent you approve adds a monitoring obligation that lands on a person, and the number of people available to monitor stays flat while the agent count climbs.

Commentary: Stanhope and Liu convert agentic AI from a capability question into a portfolio question, and the portfolio question has an answer a CMO can write down: this many agents, these owners, this review cadence.

Fix the manager layer that has to make the call

Marketing Week's Secret Marketer published Marketing needs to solve its middle management problem on September 9, and the reported incident is the most useful thing published this week.

A team ran a simple experiment: marketers brief content producers directly, with no agency choreography and no set-piece creative reviews. A junior marketer produced a brief with AI in minutes. The brief was immaculate and empty. No customer segment, no insight, no articulated role for the brand, no choices made.

The manager's feedback went to the headline, the image, and the format. The columnist's read is that the junior marketer performed exactly as expected, and the manager revealed the actual problem. That manager had built a career on channel specialism, and the people promoting them had valued fluency over framing and optimization over discernment. Execution was the only ground they felt safe on.

The line worth carrying into your next planning meeting: AI is not collapsing junior roles so much as exposing that marketing stopped being strategic long before AI arrived. Speed removed the hiding places.

Set that beside the Forrester finding. Choosing which eight agents to run demands framing, segment definition, and a willingness to say no in front of peers. A manager fluent in paid social and untrained in judgment will approve the list. Approving the list looks like leadership and costs nothing to defend.

It reminds me of what Lisa Avvocato, CMO at Vitam, said when I interviewed her on The Agile Brand podcast: "Marketing is eventually going to become this blend of creativity and growth architecture. The best marketers are not going to be choosing between storytelling and structure; they're going to be designing campaigns and strategies that work simultaneously together. If you're not learning that now, you will be left behind."

The columnist's prescriptions are concrete and cheap. Reinstate live brief reviews where people sharpen the problem. Make managers accountable for coaching strategic choices, and grade them on that. Write standards for what a good brief, a good insight, and a good creative argument look like. Protect time for practice.

Every one of those is a capability build with a named owner and a review cadence, which makes it fundable. In my framework, goals sit above the pillars, and the direction a human brings to the system stays a human act. Book 2 states it as a division of labor: analysis tells you what is and what is likely, and deciding what the organization should do is a normative act that stays with people. Marketing's middle layer is where that normative work happens at volume, and most organizations staffed it for descriptive work.

Commentary: The Secret Marketer supplies the mechanism behind every stalled AI program in the other three pieces, which is that the selection decision arrives at a layer that was never developed to make it.

Draw the line around the journey, not the function

Chung, Jean-Christophe Fann, Candace Lun Plotkin, Jennifer Stanley, and Maria Valdivieso published AI Is Blurring the Line Between Sales and Marketing in HBR on September 4. Their argument: agentic AI has dissolved the working boundary between the two functions while most companies continue to deploy the technology separately inside each one. Shared data and agents let marketing and sales activity run as a continuous loop. Capturing that requires rethinking workflows, metrics, incentives, and organizational responsibilities alongside the technology.

Read that as a constraint on the Forrester portfolio decision. If you cap the agent count and draw the cap around marketing, you have optimized one half of a loop and left the handoff running on the old seams. Three of your eight agents will duplicate work three of sales' agents already do, on a different copy of the same account history.

Josh Stephenson's September 8 report on Merlin Entertainments shows the same shape from the practitioner side. The attractions business revamped its search strategy and grew AI Overview visibility at six times the benchmark rate, under a headline quoting the operator directly: all teams need to think about this. Generative engine optimization sits nowhere on a channel org chart. It draws on product content, technical site structure, PR, and support documentation, and it belongs to whoever owns the answer a machine assembles.

This is the horizontal capability argument I am building on top of Coherence at Velocity. Intelligence, operations, and experience run across every team because those three are genuinely shared, and no function owns any of them outright. Draw your agent portfolio around functions and you inherit the org chart. Draw it around journey stages and you get a portfolio a CFO can follow, because customer journey orchestration already gives you the stages and the handoff points.

It reminds me of what Raj De Datta, CEO and Co-Founder at Bloomreach, said when I interviewed him on The Agile Brand podcast: "When you structure your team around the customer, not the channel, loyalty becomes a natural byproduct." Agile Brand Principle 5 says the same thing about mediums, and it applies to agents with more force than it ever applied to campaigns, because an agent inherits the boundary you assign it and holds that boundary at machine speed.

Commentary: Chung and his coauthors identify the seam where an agent portfolio drawn on functional lines quietly fails, and the fix costs organizational capital, which is why teams keep deferring it.

Read the seniority gradient in your own brand data

Rogers published Over half of B2B firms to ramp up brand building on September 8, drawing on Marketing Week's State of Brand in B2B survey of 300 respondents. Headline finding: 58.4% increased their focus on brand building over the past year, against 9.4% who deprioritized it.

The number to sit with runs down the org chart. Seventy-one percent of B2B CMOs believe their business sees the value in brand marketing. At the marketing director and vice president level, that figure drops to 54.5%, and 42.9% of marketing directors say their business places no value in brand at all.

That gradient describes the same layer the Secret Marketer wrote about. The people closest to execution read their organization's commitments differently from the people setting them, and the gap runs 17 points wide.

The horizon data completes the picture. Long-term strategy is the primary focus for 8.5% of these firms. Nearly 30% orient entirely toward targets that pay back inside six months.

Apply that to your agent portfolio and the arithmetic gets uncomfortable. Capability builds pay back over two to three years. Judgment builds pay back slower still. A manager layer that reads the organization as brand-indifferent and works on a six-month clock will select the eight agents that produce visible output this quarter, every time, and will be right about their incentives.

Commentary: Rogers gives you a diagnostic you can run internally in a week, which is asking the same brand-value question at three levels of your organization and measuring the spread.

Run the math on a portfolio you recognize

A $900M B2B industrial manufacturer runs 47 people in marketing and 180 in sales. Over 14 months the two functions stood up 31 AI agents. Marketing runs 19, sales runs 12. Nobody built the list. It accumulated through platform renewals and team initiative.

Fully loaded cost per agent runs about $34,000 a year: the license increment, integration maintenance, monitoring, prompt and evaluation upkeep, and the manager hours spent reviewing output. Thirty-one agents cost $1.05M. Nine have a named owner.

Three of those agents summarize account activity. One serves marketing's lead scoring, one serves the SDR team, one serves field sales. Each reads a different copy of the account history, and the three produce conflicting summaries roughly a fifth of the time. Sales stopped trusting marketing's version in month four and now rebuilds account context manually before every call, which costs the team an estimated 2,100 hours a year.

Now redraw the portfolio around five journey stages: discovery and answer visibility, evaluation and technical validation, quote and configuration, onboarding, and expansion. Eight agents cover all five. Cost drops to $272,000. Direct saving, $782,000.

The handoff saving is larger. One account-context agent serving all three teams eliminates the reconciliation work and recovers most of the 2,100 hours. At a blended $95 an hour, that is roughly $200,000, and it removes the reason sales distrusts marketing's data.

Then look at who makes the selection. The cross-functional group convening to choose the eight comprises five directors. Four came up through paid search, email, events, and ABM display. One has defined a segment and defended it to a general manager. That ratio is the finding from the Secret Marketer's column expressed as a staffing fact, and it predicts the outcome better than the scoring matrix does.

Fix the ratio before you run the exercise. Put two people in that room who have owned a P&L line or a segment definition, give the group a hard cap of eight, and require each director to write the argument for one agent they are cutting. The savings are real either way. The portfolio holds up over three years only if the people choosing it can defend a subtraction.

Do these four things before Q4 planning closes

Set the agent cap in writing and defend it. Pick the number, name an owner per agent, and set a review cadence. Forrester's finding is that resources bound the portfolio. Your finance partner will fund a bounded portfolio and will not fund an open one.

Draw the portfolio on journey stages. List your stages, map each candidate agent to one, and delete the duplicates the functional view was hiding. Agents that span the marketing and sales handoff belong to the stage, with one owner across both functions.

Audit the selection room before the selection. Count how many people in it have defined a segment, defended a subtraction, or owned a P&L line. If that count is under half, change the room. The scoring matrix will not compensate for who is reading it.

Run the brand-value question at three levels. Ask your CMO, your directors, and your managers the same question about whether the business values long-term brand building. Rogers found a 17-point spread. Measure your own, because the spread tells you how your middle layer will vote when you ask them to cut.

Name the eight agents. Then name the person who picks them, and look hard at what that person learned to see. The selection is where AI value gets made or lost this year, and no framework, matrix, or vendor roadmap will make it for you.

More than 900 conversations with senior enterprise leaders on The Agile Brand podcast have taught me something the Secret Marketer's column named exactly. Judgment moves by proximity. You learn what good looks like by sitting near someone who has it while they decide, and you learn most from watching what they leave out. That apprenticeship largely stopped happening in marketing over the last two decades, and AI did not cause it. AI made the absence visible in a single meeting. The organizations that rebuild it will be choosing eight good agents in a year while their competitors are still running 31.

Featured this cycle:

1. AI Agents For Marketing Are Here: Are You Ready?. Joe Stanhope and Jessica Liu, Forrester, September 8, 2026

2. Marketing needs to solve its middle management problem. Secret Marketer, Marketing Week, September 9, 2026

3. AI Is Blurring the Line Between Sales and Marketing. Doug J. Chung, Jean-Christophe Fann, Candace Lun Plotkin, Jennifer Stanley, Maria Valdivieso, Harvard Business Review, September 4, 2026

4. Over half of B2B firms to ramp up brand building. Charlotte Rogers, Marketing Week, September 8, 2026

Also referenced: 'All teams need to think about this': How Merlin Entertainments is approaching GEO. Josh Stephenson, Marketing Week, September 8, 2026.