Future of Marketing briefing: The agent era on training wheels

Think of the Slack bot that watches a campaign channel and turns discussion into Asana tasks; the Notion agent that assembles a clean recap deck after a client call; the reporting bot that posts daily performance summaries and flags campaigns slipping below benchmark; the lead-routing agent that enriches a form fill and delivers it to sales with context or creative platforms that generate dozens of ad variations and slot them into a preset launch flow. They streamline hand-offs and documentation but they’re still executing instructions not making choices. 

Kochava has been watching this dynamic closely. The ad measurement company’s new product StationOne isn’t pitched as a leap into autonomous marketing so much as a détente: an app that lets teams plug their data tools, ad platforms and LLMS into one interface. So instead of every employee prompting differently, StationOne standardizes the prompts, connects to systems like Slack, Salesforce or Kochava via MCP, and turns those interactions into repeatable workflows.

In practice, StationOne becomes a place where teams organize and run the smaller, task-specific agents they already have, said CEO Charles Manning. It doesn’t replace the human judgement around planning, creative direction or negotiation. It just makes the hand-offs smoother and the work faster. 

This is the real shape of the agent era right now. Not autonomy. Acceleration. The middle chapter where AI handles the work around the work, and the industry slowly decides what it’s willing to hand over next. Which is why the notion of a true agent – software that can reason, choose and act on its own – remains aspiration. Even the definition shifts depend on who’s talking. 

“Everyone is calling things ‘agents’ that aren’t agents,” said David Mainiero, chief AI officer at AI Digital, an AI-native media consultancy that helps brands and agencies running ad campaigns. “Technically, an agent is something that reasons, decides and acts across systems with limited or no human intervention.”

And that’s ok. In fact, it’s expected at this stage. Agents scale by adding capability one piece at time. Push too far too soon and, like people, they break. 

Converge is a case in point. Instead of one omniscient agent making media buying-buying decisions, it uses “tens” of narrow agents in the milliseconds before a bid, said CEO Ian Maxwell. One scores whether the impression is even worth considering. Another evaluates page or video context. Others pull external signals to infer audiences. Those pieces are stitched together and passed back to agency investment teams. 

Or to put it another way, the agents don’t choose the media. They just give the buyer a clearer picture of what they’re choosing between. 

“The ultimate aim with anything agentic is that you have agents training agents,” said Maxwell. “You have to teach one to do its job reliably before you can even think about merging them into something bigger.”

It’s similar philosophy at Immediate Media. Its sales team is already on version two of an agent built on top of its first-party data stack PRISM. It pulls together audience and contextual segments, traffic patterns, performance signals, and a record of past pitches and post-campaign reports. Reps ask questions in plain language to assemble plans, compressing what used to be a 48-hour analyst loop into minutes. It’s live and being used. 

The next phase is about feeding PRISM more research inputs, including synthetic studies, to improve pre-sales depth. Its limits are clear. It handles the high-volume, repeatable planning work around display and proposal building. Anything that draws on editorial tone, talent, brand identity, client prioritization, negotiation, or strategic judgment stays human – and will for a while. 

“A huge portion of our revenue is tied to our brands and the content we create with partners,” said Mario Lamaa, the publisher’s managing director of data and revenue operations. “That’s not something an agent can deliver. There’s still a lot of judgment, taste, and collaboration involved.”

Whether that stance holds is what the industry is now testing. WPP and Publicis are hiring agent developers to automate more of client delivery. News Corp is evaluating whether agents can eventually sell media directly. Brands are experimenting in-house. Platforms and ad tech vendors are racing to automate everything they can. None of these moves reset the market but they are already moving workflows, headcount and money. 

“You need leadership willing to experiment because some projects won;t deliver the return straight away,” said Lee McCance, chief product officer at data management business Adverity. “But the worst thing you can do is stop. The companies that persist and learn are the ones that will benefit when the next phase arrives.”

Until then, agents are raising the floor, not the ceiling. The ceiling still belongs to teams that the people who know how to use the tools rather than the tools themselves. That dynamic is especially stark for publishers. As Permutive CEO Joe Root notes, the question  isn’t whether they can build agents, it’s how they can use them consistently to rival the performance of Facebook and Google. If they can, agents claim more of the buy, If they can’t, the spend stays put. 

No prizes for guessing where his loyalties lie. Half of its publisher clients run their direct-sold campaigns through its agents today. Those agents use supply-side signals plus outcome data to predict which audiences and placements will drive downstream performance. Advertisers can also use these agents on the buy side to assemble curated audience packages across publishers. 

“Most agencies are still in the data foundations and human-in-the-loop stages,” said Root. “No one has released fully autonomous agents yet, though it’s clearly the roadmap.”

Bottom line: this phase isn’t about agents taking over. Its about teams learning to work alongside systems that are fast, confident and occasionally wrong. The leverage comes from knowing the difference. — Seb Joseph

Numbers to know

10%: Sponsored Shorts accounted for less than 10% of sponsored videos on YouTube in H1 2025, though is expected to rise substantially next year.

$750 million: annual run rate for Snap’s Snapchat+ subscriptions service, announced during the platform’s Q3 earnings

20%: Percentage by which Pinterest’s shares dropped, after reporting its Q3 results that missed earnings per share

$70 billion: The projected revenue amount Anthropic expects to make in 2028

Snap stock soars on Perplexity partnership, revenue growth

Snap’s shares rose more than 20% following the company’s earnings report, where leadership confirmed a 10% revenue growth, according to The Information. Other news which appeared to go down well is the announcement of the platform’s partnership with Perplexity, which sees the AI platform pay Snap $400 million over 2025 via cash and equity.

How OpenAI uses complex and circular deals to fuel its multibillion-dollar rise

While at the Abilene, Texas, site of a data center OpenAI is building, the platform’s chief Sam Altman recently said that tech revolutions aren’t just driven by tech, they’re also driven by finding new ways to finance them, according to New York Times.

Microsoft’s agent platform play

During its GitHub Universe developer conference in San Francisco on Nov. 4, Microsoft announced that GitHub wants to be the central platform for AI coding agents, according to Sources’ Alex Heath.

What we’ve covered

Netflix’s ad boss on the next phase, and how Amazon accelerates it

As the streaming giant marks three years of its ad business, it’s no longer talking like the new kid on the block. These days, platform execs are talking confidently, expecting to be on every major media plan.

After early, success, the NFL plans more creator-led broadcasts

Having hired four creators to host alternative broadcasts on YouTube for this season’s opening game, the NFL’s svp of social, influencer and content marketing Ian Trombetta told Digiday that the league is already determining who it’s eyeing up next.

Pitch deck: How Amazon plans to turn Q3’s $17 billion ad haul into Q4’s next big DSP push

In order to pull in more ad dollars during the holiday season, the ecommerce giant’s pitch deck leans on a familiar message: its DSP isn’t just for buying its ad inventory. It wants to be the biggest DSP for buying across the open web.

Brands set to cut open web display spend 30% in response to AI search

Advertisers may respond to zero-click search adoption by cutting their display investments with publishers on the open web by 30% in favor of CTV and paid social in 2026, per analysis by Forrester.