The media buyer of 2030 trades execution for judgment

What if the job title "media buyer" doesn't make it to 2030? "I don't think 'buyer' will be the right word anymore. By 2030, nobody will get paid to buy – they'll get paid to decide what to buy, and to check the machine understood correctly," said Rafael Lasnier, head of programmatic at agency Jellyfish. That view is gaining ground across the industry. The job is shifting from execution to judgment.

Automated setup, reports ready before you wake up, micro-adjustments handed to agents – much of what fills a media buyer's day today may not take much time at all by 2030. Nobody Open Garden interviewed seemed attached to reconciling export files. The disagreements start once you ask what replaces that work.

Ask what a buyer freed from operations actually does, and several different jobs come into view. A "tool designer," said Emmanuel Crego, managing director of French media agency Values.Media. An "agent architect," said Camille Quiqueret, head of programmatic and social ads at agency 79. Erwan Lohezic, managing director of agency Biggie France, described "a more technical profile, able to translate the advertiser's business into usable data." Eric Boyer, managing director of French media agency group CoSpirit, focused on the moments when an expert has to override the machine's recommendation.

The visions overlap on one requirement: buyers will need to understand far more than what they get a machine to execute. That leaves agencies a few things to sort out before 2030 – training, organization and fees, for a start.

Agentic tools could change how buyers work with their tech. Until now, they mostly needed to know a tool's features, settings and limits. Next, they'll also have to decide how those tools work together. What data should a buying agent use? Which actions can it launch on its own? When does it need a human sign-off?

"Where the media expert used to be a good user of highly standardized tools, they become partly the designer of those tools, because the agents keep evolving," Crego said. The work starts with a fairly unglamorous task: mapping out how you actually work. That means separating repeatable rules from questionable habits, and both from decisions that deserve a conversation with the client.

Quiqueret pictures an agent that picks partners from a brief, proposes a media plan, preps the setup once it's approved, then surfaces optimization ideas. It's up to the buyer to set the permissions and checks. Small, standardized accounts will be easier to hand over than large ones juggling multiple partners and tests.

That shift won't turn every trader into a developer. But traders will need to be able to spell out a technical need. Crego pointed to partnerships with ad sales teams and data vendors: "identifying a signal that would improve buying, an analysis you couldn't produce before, then the connections needed to get there." Not every team is ready for that kind of creative work. Most have spent years mastering the tools of a single channel.

"The media buyer could become a true media solutions facilitator," said Julien Domingues, head of trading at dentsu France. That means someone who can pull together strategy, data, technology and measurement to solve an advertiser's problem. The agency's advice would extend to choosing infrastructure and connecting it to AI.

Lasnier applies the same logic to what agencies hand their clients. The media plan would make more room for a set of instructions fed to the agents – goals, KPI hierarchy, exclusions, product priorities, seasonality, legal constraints and performance history. "A brand constitution, living and versioned, machine-readable." Advertisers would then sign off less on a fixed annual split between channels and more on a framework within which allocations can shift.

Opinions converge on autonomy once real budget is on the line. Quiqueret reserves fast adjustments for agents and leaves medium- and long-term direction to people. Lohezic wants a human check before recommendations go out to the client. Crego keeps people involved in budget orchestration, goal prioritization and success KPIs.

"You can let a machine make thousands of micro-decisions, but probably not commit several hundred thousand euros on its own," Boyer said.

Oversight will have to run on thresholds and alerts. A bid tweak and a major budget shift can't go through the same approval process. But "nobody can seriously sign off on a thousand decisions an hour," Boyer added. A human clicking "approve" on every proposal without actually reviewing it isn't much of a check.

Part of the problem is what the agent doesn't know. "An advertiser's constraints can also be implicit, harder to model," Lohezic noted. How well a brand fits the media running it can matter as much as a cost gap. Crego also pointed to annual negotiations and one-off requests around placements, priorities or targeting.

The brief fed to the machine will need to capture client knowledge that today sometimes lives only in a conversation, or in an account director's experience. Without it, an agent can produce a recommendation that's perfectly consistent with its instructions and poorly suited to the advertiser.

That's where Boyer locates some of the value people add: handling an exception, factoring in an outside event, testing an idea that past results don't fully justify. Or changing the goal altogether. A machine that keeps optimizing a KPI that no longer matters can work very efficiently on the wrong problem.

Lasnier draws that line a different way: reversibility. Bids, allocations and deal bookings look largely automatable to him when they can be corrected quickly. But the trade-off between growth and profitability, and the level of risk a brand accepts, are calls the agency has to own and defend to the client. That autonomy also depends on being able to trace the instructions and data behind every decision.

"What data can an agent access? Up to what amount can it commit budget? Which decisions need human sign-off? How do you track its actions?" Domingues asked. Those are all rules to set before handing off control. Oversight won't stop at checking whether a campaign is spending correctly – it will mean verifying the system is staying within its authorized scope.

Everyone wants buyers to understand the business. Lohezic put a fairly concrete spin on it: a conversion shouldn't carry the same value depending on whether the client wants better margins, more volume or to clear inventory. That priority needs to be translated into the data and instructions that feed the buy.

"An AI is only as smart as the data you feed it," Lohezic said. That takes skills beyond advisory work: collection, taxonomy, matching data from multiple sources, handling privacy. Buyers will need to track how a business goal turns into an optimization signal, then check what the algorithm did with it.

Quiqueret sees the agency's own signals as a way to set its agents apart. Boyer pushed back slightly: an agency doesn't necessarily need to build its own activation data. Knowing which data to use, how to combine it, keep a history and structure it properly can already set an agency apart. The test will be showing how that data improves decisions.

But feeding an algorithm well isn't enough if you're asking it to optimize the wrong outcome. "You also need to avoid confusing optimization with steering the business," Boyer warned. A falling CPC or a better cost per lead doesn't, on its own, say whether the trade-off is good for the brand. He flagged the risk of cutting brand investment because its contribution is harder to prove than that of performance campaigns. Over time, that starves the very thing that feeds performance.

The question will still be live in 2030: would sales credited to advertising have happened anyway? Does a strong ROAS justify putting another euro in the same spot? What effect are you expecting in three months versus a year? Incrementality measurement and testing will have to separate what actually drives extra results from what just claims credit for a sale.

"Agent autonomy isn't limited by AI, it's limited by measurement," Lasnier said. Feed an agent a last-click signal or a conversion attributed after exposure, and it mostly ends up amplifying a bias – faster, and with more budget behind it. So tests and measurement methods need to be solid before an agent gets more autonomy.

"Once every platform and every agent is good at optimizing a KPI, human value will mostly come down to deciding whether you're optimizing the right KPI, with the right signals and the right measurement method," Domingues said. Buyers will also need to understand what data feeds an agent, which tools it can reach, what permissions it holds and how to challenge its recommendations.

Quiqueret also noted how hard it is to agree on a shared baseline as more and more data gets modeled. Automated reporting won't settle that disagreement. It could even bury it under a polished report. The time saved on building spreadsheets has an obvious use here: checking sources, building tests and understanding the gaps. Provided that time hasn't been cut along the way.

Agentic tools could help undo an organizational structure the market already knows is flawed. "We talk about Total Video constantly, and it ends up being a pile of operational silos," Crego said. TV, programmatic and social all buy video, but rarely with the same experts, tools or data.

Quiqueret is betting on unified TV and digital buying by 2030. Lohezic pictures buyers who can follow several platforms, shift budgets around and get branding and performance working together – "multi-specialists." That profile will need to know each environment well enough to push back on its recommendations, while still thinking beyond their own channel.

That depends on the tools actually talking to each other. "Automating a fragmented ecosystem doesn't produce a simple ecosystem – it produces a fragmented, automated one," Quiqueret warned. If every ad sales team builds its own agent with no real connection to the others, the buyer will have just swapped ten interfaces for ten conversations. The payoff will be less obvious than in the demo.

There's also a bias to watch for: an agent can only weigh the environments it's connected to and the features it can access. Its best plan may really just be the best plan within its own technical footprint. Knowing what sits outside that footprint, maintaining relationships with ad sales teams and testing new options will stay a buyer's job.

"A Google agent is an excellent Google agent," Lasnier said. It's on the agency, he argued, to build the connections that let clients weigh those environments against each other. That technology layer will have to prove its worth beyond what each platform already offers.

Creative will also become a bigger part of the job, without turning the media buyer into a creative director. Quiqueret draws a line between coming up with ideas and activating or testing platforms' creative features. Buyers will need to connect message, placement context and performance more closely – which also means breaking down some walls between media and creative teams.

By this point, the 2030 media buyer is starting to look like a demanding job posting. Media expertise, technical fluency, business understanding, the ability to advise a client: Boyer and Quiqueret both picture more experienced teams, with more data and development skills built in. Quiqueret likens them to consulting teams.

Does that mean far fewer buyers? Boyer rejects the shortcut of assuming headcount falls in line with the share of tasks automated. Problem cases can eat up a disproportionate amount of the work. And teams could finally dig deeper into analysis, or test scenarios they don't have time for today. He expects some jobs to disappear, but won't put a number on the net change without knowing what new work shows up.

"I don't believe in headcount cuts. I believe in a massive transfer of skills and a full reshuffling of teams," Lasnier argued. He expects more advisory, measurement and engineering work to explain systems' decisions and take responsibility for them. That's a bolder call than Boyer's: he leaves the balance between lost jobs and new ones open.

"Teams should get leaner on pure execution, but more specialized in strategy, data, technology, AI, operations and measurement," Domingues predicted. At dentsu, he said, that upskilling runs through a head of AI and tech and a cross-functional AI community tasked with tracking experiments and sharing best practices.

The more immediate problem may lie elsewhere. "How do you train tomorrow's seniors if some of the tasks juniors used to learn the job on disappear?" Boyer asked. Language models can help with training and passing on expertise. They don't fully replace the experience of a campaign going sideways, a setup that needs fixing, or a result you're struggling to explain.

"Someone who doesn't know the channel well enough won't be able to tell if their agent's recommendation is excellent, average or completely off," Boyer added. Agencies will have to figure out how to keep that hands-on learning alive in a job that involves less execution. Otherwise, they risk spending years hunting for senior talent they stopped training.

Will advisory work replace buying on the invoice? The answers are less uniform than the talk about automation suggests. Quiqueret considers volume-based fees outdated and favors a mix of billable hours, business results and technical fees. Crego expects advisory work to take up a bigger share of the fee – still billed by the hour or as a share of budget – as buying fees potentially shrink.

Lohezic also expects technical fees for reporting, cross-platform optimization and data prep. Boyer thinks flat fees and day rates are here to stay. "When you hire a good lawyer, you're not just paying for the two hours they spend on your case – you're paying for the years of experience that let them give you the right answer in two hours," Boyer said.

Lasnier takes a sharper stance: he predicts the end of time-based billing and volume commissions, in favor of a technology subscription, expertise fees and a variable component.

"You can't preach efficiency and get paid by volume," he said. That model still depends on agreeing on what the agency actually delivered – Lasnier ties performance-based pay to measurement that's independent, shared and accepted by both sides.

Performance-based pay doesn't solve everything. Boyer warned that a bonus tied only to cost per lead or short-term ROAS can push toward exactly the trade-offs an agency should avoid. It's hard to defend brand investment when your own financial interest pushes you toward the conversion you can prove right away.

"If it all comes down to productivity gains with no extra value added, then yes, agency fees can be challenged, because the time saved only benefits the agency," Crego warned. Clients probably won't ask how many agents worked overnight. They'll want to know what that made possible with their budget – and what their agency contributed to get there.