Who owns agentic workflows? Agencies struggle to govern new tools as marketing budgets surge

"Deploying AI agents into workflows has become table stakes. Who owns those workflows, however, remains fuzzy. Agency execs argue the need for industry-wide standards and guardrails to prevent AI agents from going rogue, especially when it comes to AI media buying. Internally, though, those execs are OpenAI ad pilots and agentic workflow best practices exist in no man's land."

"Execs say they're in the process of providing guardrails around their own internal systems to prevent over reliance on generative AI for things like presentations or client insights, and vet agentic workflow tools. The topic came up during Digiday's Programmatic Marketing Summit in a town hall session, in which attendees are granted anonymity in exchange for candor."

"One brand marketing exec cited copy and pasted AI summaries in places like presentations and client insights as "bad behavior." Another said their internal teams are uploading datasets into LLMs and passing along "surface-level" insights to leadership, "completely missing the whole purpose of what we do at this point.""

""It's about us setting a tone for those who don't want to do that [copy and pasting], who do want to thrive, and then incentivizing that, and then nourishing those team members," said one of the brand marketing execs. Another agency-side exec in the town hall said they were put in charge of an OpenAI ads pilot simply because they understood agentic ecosystems."

Deploying AI agents into marketing workflows is common, but ownership and governance remain unclear. Agency leaders call for industry-wide standards and guardrails to stop AI agents from going rogue, particularly in AI media buying. Internally, teams are building guardrails to reduce overreliance on generative AI for tasks like presentations and client insights, and they are vetting agentic workflow tools. Some brand and agency managers police agentic workflows case by case. Examples include copy-pasted AI summaries treated as bad behavior and teams uploading datasets to LLMs and delivering surface-level insights that miss marketing goals. Leaders emphasize setting expectations, incentivizing desired behavior, and supporting team members who follow best practices.

Read at Digiday