CANNES, France — Artificial intelligence may be able to create a media plan, generate thousands of ads and summarize a meeting before anyone has found the conference room. That does not mean it always should.
As agencies and brands race to adopt agentic AI, the industry needs to move beyond isolated experiments and build systems that improve actual business outcomes, executives from Basis and PHD said at the Cannes Lions International Festival of Creativity.
Mike Olson, executive vice president of client development at Basis, and Emily Costello, head of integrated investment at PHD, discussed how autonomous AI tools could reshape media planning, activation and measurement. They also warned that the technology needs guardrails, human judgment and clearly defined goals.
The discussion, titled “The Agentic Shift: Scaling AI in Media Strategy,” was moderated by Zach Rodgers of Sensical Consulting.
AI climbs the maturity ladder
Costello described agentic AI adoption as a three-stage maturity model.
The first stage consists of functional agents that handle individual tasks. These may include summarizing briefs or making parts of a workflow more efficient. Connected agents represent the second stage, in which signals begin informing different pieces of the media cycle.
The third and most advanced stage is what Costello called a “horizontal orchestration layer.” At that point, multiple autonomous systems work together across functions instead of waiting for humans to pass work from one department to another.
Most organizations remain somewhere between the first two stages, she said. The larger opportunity will arrive when agencies achieve “true autonomous synergies across workflows.”
In other words, the robots have learned a few useful chores. They have not yet been handed the keys to the agency.
Olson said agencies are moving at different speeds. Major holding companies have sophisticated AI platforms, while independent agencies may lack the data and technology needed to build their own systems.
Basis is seeking to fill that gap with agentic capabilities spanning planning, strategy and activation. The challenge is not simply gathering more information. Agencies must determine what data will help produce the outcome they actually want.
Jurassic Park had an AI strategy
The abundance of potential AI applications has created a familiar technology problem. Companies can build almost anything, but they may not have stopped to ask whether anyone needs it.
“There’s so much going on right now that it’s kind of this ‘Jurassic Park’ model,” Olson said. “They never stop to think if they should just because they could.”
Before deploying an agent, marketers should agree on the goal and the language used to define success, he said. Otherwise, a project can head in several directions without becoming useful to the agency or its clients.
Costello offered a similar warning about creative testing. AI can produce more ads and experiments than human teams could manage on their own. Still, more variations do not automatically produce more insight.
“You need to remain as focused as possible when it comes to creative testing,” she said. The aim should be to identify what moves the needle, not to conduct so many experiments that the results become noise.
Guardrails without an AI hall monitor
PHD wants employees to experiment with AI without feeling that every prompt is being watched by a compliance officer carrying a clipboard.
The agency’s Omni AI platform provides a protected environment with controls already in place. That lets teams test ideas while maintaining safeguards around client data, Costello said.
PHD is not including individual AI usage in employee performance reviews. Costello said that kind of mandate could create fear, especially among workers already wondering whether the efficiency they produce will eventually make their roles redundant.
Instead, the agency is emphasizing how automation can free people to spend more time on strategic work.
“When you start to see capacity release, what that enables is for us to start to invest time,” Costello said.
Basis also describes itself as an AI-first company. Olson said employees are encouraged to build tools while respecting rules governing sensitive information and the large language models they use.
In one example, Basis’ head of revenue operations used Claude to build a dashboard connecting internal tools for business planning. Basis is moving that application into its own technology infrastructure so the company can support it.
That is vibe coding’s corporate graduation ceremony: one day an employee is experimenting with an AI tool and the next day IT has adopted the offspring.
Beware the slop machine
AI can increase productivity, but it also can turn a simple thought into a novella that nobody requested.
Olson described the downside as “all the slop.” A matter that once required a 30-second phone call or a paragraph can become a long document that colleagues must read and verify.
“Unless you can sign your name to it and you know every word it produced, don’t use AI just to produce a bunch of content,” he said.
Costello agreed that agencies need new quality-control systems. Employees must develop the habit of checking AI-generated work to ensure it solves the original problem without creating several exciting new ones.
AI has therefore “created a premium on humans,” she said. People are still needed to review the work, exercise judgment and catch the confident nonsense before it reaches a client.
Rodgers noted another concern. Junior employees traditionally develop judgment by handling lower-level assignments. If AI automates those jobs, agencies will need another way to cultivate future leaders who can oversee the machines.
Olson said Basis encourages employees to share AI tips with colleagues. Younger workers sometimes discover effective techniques but keep them to themselves as a personal advantage. A culture of sharing can spread productivity gains while reducing fear about the technology.
Sustainability waits in the lobby
Agentic AI also carries financial and environmental costs because its models require substantial computing power.
Costello said agencies must balance fiscal responsibility with experimentation. One approach is to build reusable skills rather than constantly creating new agents, reducing repeated token and computing costs.
Olson compared the debate with the advertising industry’s earlier focus on reducing carbon emissions from programmatic media. Agencies and brands asked vendors about sustainability while also demanding more targeting, more inventory and global reach.
“You can’t have both at one time,” he said.
AI capabilities are currently taking priority over environmental concerns, Olson observed. He expects sustainability questions to return as token costs fall and AI tools become more established.
For now, the industry appears to have placed its carbon calculator in a safe location where it cannot interfere with the product roadmap.
Outcomes still pay the bills
For all the enthusiasm around AI, Olson said the basic purpose of advertising has not changed.
“We’re in advertising, right? I mean, brands want outcomes,” he said. “At the end of the day, if I run a furniture store, I wanna sell more furniture.”
AI is worthwhile when it makes agencies more efficient, improves service or helps clients reach their business goals. It becomes less useful when it adds fragmentation and operational headaches.
The technology also may broaden the meaning of creative personalization. Olson pointed to Coframe, a company that can alter a website for each visitor. That approach extends optimization beyond the ad and into the landing-page experience.
Costello said consumers increasingly expect personalized experiences, although marketers must avoid crossing the line from relevant to creepy. AI may create many capabilities that the industry has not considered yet, she said.
Looking ahead, both executives named orchestration as the priority.
Costello expects AI systems to connect strategy, activation, measurement and governance without relying on today’s linear handoffs. Olson said Basis wants that orchestration to extend from planning through billing reconciliation.
“If we can have agentic solutions helping solidify and streamline that to create autonomous processes, that’s what we want,” Olson said.
The machines may soon plan the campaign, activate the media and reconcile the bill. Humans will still be needed to decide whether any of it was a good idea.
