Your agency is selling you an agent and a governance deck

Happy Thursday.

Count the agentic announcements from one week of Advertising Week. Dentsu's Carat is launching a planning, buying and optimization agent with NBCUniversal, FreeWheel and an AI firm called Newton, with one luxury retail client nobody will name.

Tinuiti opened its measurement layer to AI agents through an MCP server. Stagwell relaunched Assembly as Stagwell Media around a system named Machine OS.

Digiday's write-up is less about the agents than about what every executive said next. Rebekah Shalit, head of platforms at Dentsu: "Every task and everything we do sits on a sliding scale of low to high human oversight." Lauren Wetzel at WPP Media: "Our whole approach is that governance and human accountability are central to agentic media buying."

David Dworin, FreeWheel's chief product officer, on why a human stays in the loop: "they're new, and we don't trust them, and they make mistakes." The pitch and the disclaimer in one breath.

Here's what I'd listen for in the next agency QBR. Wetzel's version has the detail: approval thresholds, and "a clear separation between what is probabilistic AI recommendations" and decisions with a financial impact. That's a real design.

So the question is where the threshold sits, in dollars, and who set it. If the agency set it, you've outsourced your risk appetite with your media plan.

Carat's Carrie Drinkwater is the bullish one, and her argument is about data: "in a year we're going to know so much about your business, your media." The agent learns from your spend. Nobody on stage said who keeps the learning.

None of this is cynical. Agencies are rebuilding their job before clients do it for them, and the governance language is them being careful in public. It's also doing something for the agency: every approval threshold is a line that says this decision was yours, not ours.

The agent will arrive in your next pitch deck with a slide on oversight right behind it. Before you nod, ask what happens the day it reallocates 15% of Q4 at 2am. Who approved the threshold that let it?

Today’s sponsor:

Peep Laja, CEO of Wynter, on the word that stopped meaning anything:

Three camps, which he credits to Anthony Pierri: Asana went all in with "The OS for human-agent teams." Drata kept its category and made AI the edge. Rillet's headline doesn't mention AI at all.

Buyers roll their eyes at the label now. If your product really has AI doing real work, the eye-roll is the problem you have to write around.

39.8%. That's the drop in organic clicks when an AI Overview appears, in a randomized field experiment on 1,000-plus Google users by Saharsh Agarwal and Ananya Sen, cited in Digital Content Next's review of the evidence. Pew saw 8% of visits click a result with an AI summary against 15% without.

DCN pulled the studies together this month; the studies themselves are older. The pattern isn't.

Our Agency Performance Review Framework, for the review that happens before the agent does:

Attach two quarters of invoices and timesheets, and get your own number before the agency's agent claims it.

The agencies understood the first half last week. The second half is the approval threshold, and it only counts as yours if you wrote it.

We provide consulting services for CMOs on AI strategy for marketing teams through our Teamless offering.

If your agency can run agents, so can your team. Three ways to try one on something small:

Appy.ai — best for teams that want a plain-English goal delegated across a set of AI coworkers, with a named lead agent you can hold accountable. Try Appy.ai here →

Genspark AI — best for a workspace that bundles agents, an office suite and content generation, for teams tired of six tools that don't talk. Try Genspark AI here →

Littlebird — best for a personal assistant with standing context across Gmail, Slack and your calendar, so the agent knows what you meant last Tuesday. Try Littlebird here →

→ We have 300+ curated, filtered, and categorized AI tools in our directory. Browse to find the best ones that help your marketing team.

Before an agency grades your maturity for you: the AI Marketing Maturity Model Framework. An assessment and roadmap from experimentation to systems that carry revenue, so the pitch starts from your scorecard.

→ We publish workshops, courses, guides, manuals, and more on AI marketing for free. Browse our learning resources for free.

Nikon's Small World contests have run since 1975 on one promise: this is what the real thing looks like under a microscope. This year's winner, a video of cilia in the airways of a child with a rare disease, was questioned by microscopists within days of the September announcement. The comments were on a Nikon LinkedIn post. The post has since been deleted.

Nikon defended it for weeks, then disqualified it because the entry "did not comply with the competition rules regarding generative AI," and moved the runner-up to first. The entrant says he rendered the regions below the cilia to "improve presentation".

The rule existed. Nobody checked until the audience did. What's the rule in your brand that nobody has tested yet?

You’re reading this, and so are nearly 20,000 of your peers. Some of them are your next customers.

If you sell to marketing teams, a single placement is €500 and a full week is €2,000.

If you want to place an ad, simply reply to this email or check our media kit.