Milena Traikovich is a leading voice in MarTech operations, specializing in the intersection of demand generation and AI governance. With a career built on optimizing lead-nurturing campaigns and performance analytics, she now navigates the complex landscape of “agentic marketing,” where autonomous AI agents handle everything from SDR outreach to brand voice. As organizations transition from simply building AI to managing a living ecosystem of bots, she provides the framework necessary to ensure these digital workers remain effective, secure, and aligned with evolving business goals.
This conversation explores the lifecycle of AI agents, moving beyond the initial excitement of deployment to the critical realities of long-term ownership. We delve into the widening gap between perceived and actual accountability, the historical lessons of software maintenance, and how to distribute governance across existing marketing and IT teams to prevent “output drift” and security risks.
Many organizations are rapidly deploying hundreds of agents outside traditional IT departments. How do you reconcile this “build-first” mentality with the long-term need for maintenance and policy updates?
It is a fascinating and somewhat chaotic time because we are seeing a massive shift where marketing teams are becoming de facto software developers. I recently spoke with a colleague at a major utility who is releasing 200 agents next quarter, and almost all of them are being built outside of IT. This is incredibly impressive from a speed-to-market perspective, but it creates a vacuum of responsibility once the “new car smell” of the launch wears off. We are so focused on the thrill of building that we aren’t asking what happens when a model updates or a brand policy changes. If the person who wrote the agent’s instructions moves to a different team, that agent becomes a “ghost in the machine” that no one knows how to fix or even shut down.
The data suggests that 70% of large enterprises are already running custom AI agents, yet there seems to be a massive disconnect regarding who actually “owns” them. Why is this clarity so hard to achieve?
The disconnect is truly striking when you look at the numbers. Research shows that while 85% of IT professionals say a named owner exists for every AI agent, only 42% feel that ownership is actually clear—that is a staggering 43-point gap between theory and reality. I think this happens because we are stuck in a tug-of-war between centralization and functional expertise. About 40% of leaders believe a Chief AI Officer should own agentic marketing, but if you ask AI leaders specifically, that number jumps to 52%. Meanwhile, marketing executives feel they should own it because a central AI group won’t understand if an agent’s tone is off or if the segment logic is outdated.
You mentioned that “governance shows up for the launch” but then fades into a quarterly rhythm. What are the immediate risks of this “set it and forget it” approach?
The most immediate and invisible risk is what I call “permission sprawl.” When organizations spin up agents, they often do so by cloning a human user’s profile to save time, meaning a campaign agent might be walking around with the full CRM access of the manager who set it up. At launch, that seems like a minor shortcut, but fast forward a few months, and you have an automated system with high-level data access that no one is auditing. Beyond security, there is the functional risk where 65% of organizations run a review before deployment, but then the agent keeps working every single day while the humans only check in once a quarter. The world moves much faster than a 90-day review cycle, and your agent can start saying things that are no longer true almost immediately.
Looking back at the history of software development, specifically the 1968 NATO conference and studies from 1980, what can we learn about the hidden costs of keeping these systems running?
History is repeating itself in a very predictable way. In 1980, a study of nearly 500 organizations found that maintenance consumed roughly half of the entire software budget, which is a figure that usually shocks modern marketers. The most revealing part of that study was the breakdown of the work: fixing actual “bugs” or defects was the smallest category. Most of the budget went to “perfective” work—where requirements changed—and “adaptive” work, where the environment around the software evolved. For an AI agent, this means the agent isn’t “broken” in a technical sense; it’s just that the pricing changed, the brand voice was updated, or the underlying model was deprecated.
Can you paint a picture of what happens to a “correctly written” agent when the world moves on without it?
Imagine your “offer agent” introducing itself at your next team stand-up. It might say, “Hi, I’m the offer agent; I was written in March against a promotion that ended in June, and I’m still sending it out to leads today.” It’s a funny image, but it’s a reality for many teams right now. You might have a content agent written against a March positioning statement, or an SDR agent using an ICP that predates a major strategic pivot. None of these are technical defects, but they are failures of maintenance. The agent is doing exactly what it was told to do, but the instructions have become obsolete, leading to output drift that can damage your brand’s credibility.
Instead of hiring new headcount, you suggest spreading accountability across existing roles. How should a marketing department divide these responsibilities?
You don’t need a massive new budget; you just need named accountability. Marketing Ops is usually the best fit for managing the instructions and the revision cadence because they already own the general workflow. The Brand team should own the “drift” question—they are the ones who can sense if the output still feels like the company. Your platform or AI team should be the ones watching for model releases and translating those technical changes into something the marketing team can actually use. Finally, retirement—the most overlooked stage—should be assigned to whoever owns the budget line, because they are the most likely to notice if a redundant agent is still burning resources.
For a team that is currently managing a small handful of agents, how do they build the right habits now before they scale to 80 or 100 agents?
Habits are significantly easier to build when you are managing eight agents rather than 80. I always tell teams to take two very specific questions into their Monday morning meetings: Who owns our oldest agent, and when did anyone last look at what it’s producing? We have the opportunity to skip the “software crisis” that the tech industry went through decades ago by recognizing right now that shipping is just the beginning. Whether you assign ownership per agent, per campaign, or per platform, the key is to move away from “fuzzy” responsibility and ensure that every digital worker has a human supervisor who is responsible for its retirement.
What is your forecast for the evolution of AI agent governance?
By 2028, we will see a shift where “Agent Supervisor” becomes a standard certification within Marketing Ops, and vendor tools will bake these lifecycles directly into their platforms. We’ll move away from manually checking for drift to having “supervisor agents” whose sole job is to audit the performance and compliance of other agents. However, the organizations that will win are the ones that don’t wait for the software to solve it. They are the ones defining the human org design today, ensuring that when an agent speaks to a customer, there is a clear line of accountability back to a person who knows exactly why it’s saying what it’s saying.
