Marketing teams are shipping AI agents at pace, but the question of who looks after those agents after launch is still unresolved. New enterprise research suggests most companies have named an owner on paper, yet fewer than half believe that ownership is actually clear.
The gap between ‘named’ and ‘clear’
Ivanti’s survey of 1,500 IT professionals found 85% said every AI agent has a named owner, but only 42% said ownership is clear. That 43-point gap is where agent deployments start to drift.
A separate Kana survey of 225 senior leaders at large US enterprises found 70% already run custom AI agents on real marketing work; only 3% run none. The build-versus-buy question is largely settled. The open question is what happens next.
Marketing and central AI teams can both assume the other is responsible
Views on who should run agentic marketing are split. About 40% of leaders point to the chief AI officer, rising to 52% among AI leaders, while marketing executives often prefer their own function or a shared model. Both positions have merit: central teams worry about data access, regulatory exposure and consistency; marketing teams know whether an agent’s offer, tone and segment logic are still correct.
The practical fix is to split ownership rather than hand the whole problem to one side. Central teams can own access, data and the model layer. Marketing can own instructions, tone and truthfulness of output. Both need names attached.
Launch governance is not daily management
The Ivanti research also found 65% of organisations run a review before an agent goes live. Governance is strong at launch, then often drops to a quarterly cadence while agents keep working daily. Permission sprawl can start immediately when organisations clone a human user’s profile to spin up an agent. A campaign agent may carry CRM access from whoever set it up. If that access becomes inappropriate later, someone has to notice.
Five questions to assign now
Start with one production agent and answer these five ownership questions:
- Who writes the agent’s instructions?
- Who revises them, and on what cadence?
- Who tracks model releases and translates changes for the team?
- Who monitors output drift, and against what baseline?
- Who decides when the agent gets retired?
Most teams can answer the first immediately; the other four are often unanswered. A workable approach is to spread these duties across existing roles: marketing operations takes instructions and revision cadence; brand monitors output drift; the AI or platform team watches model releases; and the budget owner handles retirement because they will notice when something is still running.
The software maintenance lesson
Software engineering learned this lesson decades ago. At the 1968 NATO conference in Garmisch, practitioners formalised a lifecycle that included maintenance and retirement. Later research by Bennet Lientz and Burton Swanson covering 487 organisations found maintenance consumed roughly half the software budget. Most of that work was perfective or adaptive — responding to changing requirements or environment — not fixing defects.
AI agents face the same forces. A content agent tuned to a March offer, an SDR agent built on an old ICP, or brand guardrails tied to a deprecated model version are not broken. They are simply out of date, and someone needs to decide what changed and what to do about it.
Salesforce has begun framing this with an agent development lifecycle and roles such as Agent Supervisor. Other vendors will follow. The tooling will arrive; the org design is the part marketing teams need to build now, preferably while the number of agents is still small.
Source: MarTech
