One pattern keeps coming up in my ad hoc conversations with engineers, designers, and business leads about AI.
People are optimizing inside their current frame: "I want it to do this for me." "I need it to solve that." Maybe you don't. The way we produce output in IT is in flux, AI-aware workflows are evolving week by week. The deliverable you're trying to create might not need to exist at all.
Chris Argyris and Donald Schön called this single-loop vs. double-loop learning.
Single-loop asks: are we doing the thing right?
Double-loop asks: are we doing the right thing?
Most AI conversations I hear are firmly single-loop: How do I prompt better? Can it generate my deliverable? Few people are questioning whether the deliverable itself still makes sense.
"We need to integrate AI into this workflow." – Does your workflow still make sense though?
"I want to migrate the design system." – Is your current system or setup still relevant to your business?
"I want to edit this prototype in my design software." – Are you sure you still need that software?
It's not the task that changes, but the work – and the workflow.
"Transcending the existing system" as Béla Bánáthy framed it – stepping outside the workflow entirely. Question your framing before you question your tactics.
To shift the mental model, we need to ask more fundamental questions – and challenge ourselves and business leads accordingly. Step back from the task and look at the goal behind it, and you'll find the task was built on an assumption that may be already obsolete in 2026.
Are you still single-looping?