Donella Meadows identifies leverage points in systems — places where a small intervention produces large change. Her most counterintuitive finding: changing parameters (adjusting numbers, thresholds, rates) is usually the least powerful intervention. Changing information flows — who gets what information, when, in what form — is structurally more powerful because it changes how the system learns and adapts, not just what it does in a given state.
I’ve applied this directly in two contexts and it’s changed how I diagnose design problems.
In the margin risk UI at SoFi, the initial intervention proposals were all parameter changes: adjust the threshold display, change the warning trigger level, modify the calculation basis. These were all reasonable. They were also all downstream of the real problem, which was an information flow problem — users weren’t receiving the information that would let them form accurate beliefs about volatility sensitivity at the moment of decision. Changing what information surfaced, and when, was more powerful than adjusting any of the parameters.
In AI literacy work, the instinct is always to add more — more workshops, more examples, more documentation. These are parameter adjustments. What actually moved adoption in my experience was changing feedback loops: versioned prompt logs that made iteration visible, shared failure documentation that made model limits concrete, output evaluation sessions where designers assessed model reliability before being told what the benchmark said. Changing the information flow changed what designers learned from their own experience with the tools.
The design implication for AI products: most intervention discussions happen at the parameter level. The more powerful questions are about information architecture — what does the user know, when do they know it, and how does that information shape their subsequent behavior and beliefs?
Slowsignal 
Leave a comment