In instrumental interaction the user controls each step; in agentive interaction the user delegates a goal

What it is

In instrumental interaction the user is the operator: select a file, set a parameter, click execute—every step is issued by hand. In agentive interaction the user becomes a principal: state the goal "compress these photos to under 2 MB," and the system decides and performs the steps. The dividing line is not the degree of automation but who holds the steps.

Why it happens

The control structure shapes the user's mental model: in tool mode the user builds a cause-effect chain from operation to outcome, and failures can be localized to a step; in agent mode the user builds a contract between goal and result, and the process is the system's job. Control moves inside the system, and the evaluation question shifts from "did I do it right" to "did it do it right."

Studying it

Use matched delegated tasks: complete the same goal via tool steps and via goal delegation, measuring process-report accuracy, recovery strategies after errors, and trust trajectories. Eye tracking and input logs distinguish step-by-step monitoring from result checking, and think-aloud protocols surface the difference between "what should I click" and "it should handle this."

Where it stops holding

Real products are hybrids: batch jobs with confirmation steps, wizards with automatic suggestions. Judge the mode by where control sits, not by interface vocabulary. For high-risk, legal, or identity-involving tasks, delegation may be rejected even when technically possible, while expert users on low-risk tasks will happily give up step control.

Applying it

• Fix the control point per feature—stepwise operation or goal statement executed by the system—and never blur the two inside one flow.
• Make delegation input an acceptance-ready goal description (object, constraints, completion criteria), not an instruction that smuggles in steps.
• After the agent finishes, show a result summary checked against the stated goal so acceptance has evidence.
• Watch whether users debug "which step went wrong" or "what did it do wrong" after failures; use that to check the product's mode against expectations.

Related

• Same group: B4.16.2 The two modes demand different feedback: tools need immediate response, agents need visible process · B4.16.3 Delegation transfers control and responsibility together; its boundary must be stated before the handoff · B4.16.4 The two modes can coexist in one product, but switching should be the user's decision, not the system's · B4.16.5 Agentifying operations users could do directly removes space for skill expression
• Nearby: L3 Intelligent System Interaction · B4.10 Direct manipulation
• Search terms: instrumental interaction · agentive interaction · delegation