In this final installment of our series on AI and trust, we examine how giving users control and agency over AI behavior helps them feel safer and more comfortable. Social science literature offers two competing perspectives on the relationship between trust and control: some researchers argue they are inversely related (the less trust exists, the more control is desired), while others contend they are complementary, with greater control in certain environments actually fostering trust development (Vosselman, 2009). At Thread, we subscribe to the latter view. By giving users the right degree of control over AI agents in Thread Service Desk and adhering to the Human-in-the-Loop design pattern, we aim to reduce the discomfort that comes with AI's inherent unpredictability.
Within the Thread Service Desk, control manifests in several distinct ways.
It first appears during client issue intake. When a client reports an issue, the Triage Agent immediately activates and gathers relevant information that a technician may find helpful. While the Triage Agent is working, the technician sees a “Take Over” button at the top of the chat window. Having the option to interrupt the agent allows the technician to feel more at ease because it gives the technician a way to course-correct from undesired AI responses and behavior.
By being able to intercept the agent at any point, the technician remains the ultimate decision maker in how their IT team serves clients. The more the AI acts in an unpredictable manner, the less a technician is likely to trust its future outputs. We give control to the technician to reduce the amount of times they experience AI’s unpredictability. Things can and do go wrong with AI; to ignore and dismiss the truth would be misleading and illusory. The “Take Over” button creates a failsafe that makes space for trust to develop, precisely because it acknowledges that things can go wrong.
Control also appears in how technicians configure the Triage Agent's behavior. Before the latest Triage Agent custom rule redesign, technicians defined agent behavior through a single open-ended prompt field where they could specify any instructions they wanted the agent to follow.
The problem with this approach is that it buries meaningful controls inside an invisible layer.
Admins may want to adjust specific behaviors, such as having the agent communicate differently with different clients, but many users will never discover that capability if it is not made explicit.
In the redesigned Triage Agent, we addressed this by surfacing that hidden capability. Client-specific configuration is now a visible dropdown selection in the UI, making the control clear and actionable.
Beyond UI configuration, the product and design team also leverages the Human-In-The-Loop principle to help technicians feel in control of their AI tools. In an upcoming feature, the team is building an agentic workflow to help technicians complete ticket-related actions, including tasks like sending internal notes, drafting customer replies, and running automations. At key steps, the AI will pause and request explicit confirmation before proceeding, giving technicians the opportunity to review, adjust, or cancel any action before it's taken.
By requesting the technician’s approval, we invite them into the AI process and emphasize the power of agency through collaboration.
Across intake, configuration, and agentic workflows, a consistent principle emerges. Trust in AI isn't built by making AI invisible or frictionless; it's built by keeping humans meaningfully in control.
Thread’s Magic Analytics, our new reporting feature, supports that by giving IT admins real data on how AI is performing in their team environment. Metrics like zero-touch resolutions and ticket volume that AI handled turn vague, unquantified results into something concrete and measurable. When we show admins exactly what the AI is doing and what metrics we used to reach that conclusion, we put the admins in charge of AI on their teams.
In the last two articles of this series, I emphasized the importance of demystifying the “black box” quality of AI operations and creating a consistent experience in cultivating trust. Transparency and consistency were two of the main things. However, neither trait holds up when users don’t feel like it was their choice to use AI. Control is the buy-in that allows trust to emerge. As AI takes on more complex tasks in Thread Service Desk, we'll continue to look for places where surfacing control, rather than abstracting it away, is what allows trust to take root.