JPMorgan's Programmatic Lead Just Told the Industry AI Needs Humans in the Lead, Not Just in the Loop. The Fiduciary Argument Is the One to Steal.
Melissa Bonnick, head of programmatic at JPMorgan Chase, used a phrase at AdExchanger's Programmatic AI conference that the rest of the industry should adopt. Humans in the lead, not just in the loop. The fiduciary framing is the one that survives a board review.
Humans need to discuss not just how to automate certain tasks, but which processes ought to be automated and why.
JPMorgan Chase's head of programmatic, Melissa Bonnick, used a phrase at AdExchanger's Programmatic AI conference last week that the rest of the industry should adopt. AI needs humans in the lead, not just in the loop.
The framing matters. Human in the loop became the lazy phrase for any AI workflow that had a person sign off at the end. Bonnick's reframing puts the human at the front of the decision, not the back. The human decides which processes get automated and why. The human sets the guardrails. The human owns the outcome.
She gave the bank's reasoning. Finance has a fiduciary duty that does not allow for we let the agent decide. Every campaign needs an approval gate before it goes live. The privacy posture is non-negotiable. AI agents can draft, recommend and pre-build. They cannot ship.
That is the argument every CMO presenting an AI roadmap to a board should be using. Not a list of what AI now does. A list of what humans decide AI will do.
Why it matters
Australian financial services, healthcare, government and education all carry similar fiduciary or regulatory exposure. The AI agents will run our campaigns pitch is going to look reckless in a 2027 board review. The AI agents will draft and humans will approve pitch is going to look like governance.
The other reason this matters is procurement. Australian regulators have started asking how AI decisions are made and who is accountable. APRA's CPS 230, the Privacy Act amendments and the ACCC's incoming digital fairness work all assume a human decision-maker behind the AI output. We let the agent decide is going to fail an audit.
Industry estimate for autonomous agent-to-agent ad trading at scale. The governance question is now, not later.
What to do about it
The play here is governance, not technology selection.
Write down which marketing decisions humans will own forever. Brand positioning, pricing, audience targeting around sensitive categories. Do not delegate these to AI.
Write down which decisions AI can make with human review. Creative variants, audience expansion, bid adjustments inside a defined range. Reviewable, not automatic.
Write down which decisions AI can make autonomously inside guardrails. Budget pacing inside daily caps, language translation inside an approved style guide. Constrained, not open-ended.
Build the approval flow before you turn the agent on. The ship-to-live button must require human sign-off for any new creative or new audience for the first 90 days of any agent.
Audit quarterly. The agent will drift. The humans need to notice.
Bonnick's framing is the one that gets agreed at the board level. The CMOs adopting it now will look like leaders. The CMOs still saying human in the loop will look like they read a vendor deck.