Tool or agent?
Make the boundary move.

Drag systems into the boundary map, then turn autonomy up and watch the consequence chain change. A tool waits to be used. An agent decides.

DRAG CARDS INTO A CLASSIFICATION ZONE3 SYSTEMS / UNSORTED

Position-based classifications and consequence labels

Read the explanation

The source stores three illustrated systems with normalized canvas positions. Coffee begins at point two six, feed point five, and banker point seven two. On pointer release it labels a card tool when x is less than point five, otherwise agent. Bars use five hundred pixels per normalized x unit. This classification is a visitor interface rule, not a scientific test of intelligence, goals or autonomy. It does not infer actual system capabilities. The slider changes one displayed consequence label. Values below thirty five are low, below seventy are medium, and seventy or higher are high. The bars mark the two thresholds at five pixels per slider point. No numeric social or economic consequences are simulated by that effect function, and it does not change card classifications. The visual rhetoric should not be confused with measured impact. Reset returns the three original coordinates and labels, giving one tool and two agents, and sets the state text to three systems unsorted. Bars use one hundred ten pixels per card. It calls the label function without resetting the autonomy slider. Exported Markdown includes current slider and consequence label plus each card classification. The source lexical variable replay shadows the replay button identifier, so its onclick assignment does not attach the intended button handler. Offline p5 absence occurs before the later control handlers; matching that failure is not proof of a healthy interactive canvas.

Agency Boundaries and Autonomy Modeling

How does the Agency Boundary Lab classify systems into tools versus agents, and how is modeled consequence determined?

The Agency Boundary Lab divides systems into tools and agents based on their horizontal coordinate on an interactive canvas. When a card is dragged across the midpoint (x = 0.5), it switches classification between a tool (a system that waits for an external trigger) and an agent (a system modeled as making autonomous decisions). Autonomy is set via a slider from 0% to 100%, which maps directly to three discrete consequence tiers: LOW for autonomy under 35%, MEDIUM from 35% through 69%, and HIGH for 70% and above.

The lab uses a stylized spatial threshold (the midpoint at 50% canvas width) and fixed numeric slider cutoffs (35% and 70%) for demonstration purposes. These cutoffs represent simplified heuristic boundaries rather than validated empirical safety risk or operational autonomy standards.

Try a worked example

Focus the Autonomy slider and press the ArrowRight key or set its value. When autonomy is adjusted upward from 35% to 70% or higher, the consequence badge switches from MEDIUM to HIGH and the autonomy percentage readout updates to match the new setting. Conversely, reducing autonomy below 35% shifts the modeled consequence readout to LOW.

Canvas Midpoint Threshold Mechanics

The p5.js canvas divides horizontal space at x = 0.5. Releasing a draggable card to the left of the midpoint assigns it to the 'tool' zone ('waits for a trigger'), whereas releasing it on or to the right of the midpoint assigns it to the 'agent' zone ('makes a decision').

Autonomy Slider and Consequence Tiers

The slider adjusts an integer autonomy percentage from 0 to 100. In the code, consequence levels are determined strictly by threshold checks: any value strictly below 35% outputs 'LOW', values from 35% up to 69% output 'MEDIUM', and values 70% and above output 'HIGH'.

Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.