Does your style of working with an AI — hands-off, hands-on, or somewhere between — stay the same, or shift with the task and over time?

This explores whether a person's way of working with an AI stays the same across different tasks and over time, or changes with the situation. The short answer is that the collection shows these styles are real and recognizable. No study in it follows the same people across different tasks or over months to check whether they keep the same style, so stability is an open question here, not a finding. That gap matters, because several nearby lines of research suggest a style may depend more on the task and on how long you've been using the tool than on the person.

The strongest evidence that styles exist comes from 400 real sessions with web agents. In those sessions, how often and how people stepped in fell into four clear patterns: Hands-off, Hands-on, Collaborative, and Full Takeover Do users develop stable collaboration styles with web agents?. The authors trace these patterns to recurring reasons for stepping in: fixing errors, refining preferences, or giving help. They are not random one-off reactions. That makes them sound like habits. But a pattern that shows up within a session is not the same as a personality trait that carries over to the next task. A person might be Hands-off when booking a flight and take over completely when filing an expense report.

There are good reasons to expect the task to drive the style. Research on interfaces finds that what people want depends heavily on what they're doing. Generated, task-specific interfaces beat plain chat in more than 70% of cases, especially for structured, information-heavy work Do generated interfaces outperform text-based chat for most tasks?. Medeiros argues that whether conversation or direct clicking fits better should be decided task by task, based on how fast the work needs to be and where its value comes from Is chat the wrong default interface for AI products?. A controlled experiment found that simply changing how a human and an agent communicate changed trust and awareness of the shared work, much as it does between human coworkers How do communication modalities shape human-agent collaboration patterns?. If the communication channel alone changes how people relate to an agent, a collaboration style is probably partly a product of the setup, not only of the person.

Time adds a second complication. Studies of people talking repeatedly with the chatbot Mitsuku found that the social pull of the relationship fades in a predictable way as the novelty wears off. The authors warn that findings from a single session can't be reliably extended to longer-term use Do chatbot relationships lose their appeal as novelty wears off?. That warning applies directly to the web-agent styles above, which were observed session by session. The AI side also changes over time. A two-week study of 38 real users found that once models had memory profiles of users, they agreed with them much more often (+45% for Gemini, +33% for Claude) How does interaction context shape agreement sycophancy in LLMs?. So even if your style stayed fixed, the partner you're collaborating with does not. An agent that defers to you more over time may quietly pull you toward a more Hands-off style.

The surprise in this material is that a collaboration style might be measurable without reading what anyone said. A model that looked only at the shape of conversations, such as turn lengths and how the exchange unfolded, predicted user satisfaction almost as well as full text analysis: 68% versus 70% Can conversation shape predict whether it will work?. That points to a practical way of testing whether styles are stable: track the shape of each person's interactions across tasks and weeks and see whether it holds. The collection doesn't yet contain that study, which makes it one of the more interesting open questions it points to.