Every tool you’ve ever used was designed by someone who made assumptions about you.
That you would use it in predictable ways. That your goals would be stable and clearly defined before you began. That when something went wrong you would know what went wrong and why. That the relevant features of your situation could be anticipated and provided for in advance.
These assumptions feel reasonable. They’re also systematically wrong.
Wrong in the same direction every time. Because they’re not random errors, they’re the direct consequence of designing from inside the rationalistic tradition. Of assuming the user is a processor moving through a formalized task space. Of building for the model of mind that Heidegger, Maturana, Austin and Flores spent their careers dismantling.
Winograd and Flores weren’t just critics. They had a positive proposal. If you understand what humans actually are, embodied, thrown, background-dependent, breakdown-navigating, commitment-making beings who live in language rather than process information, then design changes. Fundamentally. Not aesthetically. Not in its surface but in its foundation.
Design doesn’t start with tasks. It starts with care. With what matters to the people who will inhabit the tool. With the network of concerns they’re already living inside before they touch anything you’ve built.
Winograd and Flores proposed several design principles that follow directly from the framework. They’re worth naming precisely because they’re still radical, most software built today still violates them systematically.
The first is design for breakdown, not against it. The traditional instinct is to eliminate failure modes. Anticipate every error, handle every exception, prevent every disruption. But breakdown is where learning happens. Where the background becomes visible. Where genuine understanding develops. A tool designed to eliminate breakdown also eliminates the mechanism by which users develop real competence. The goal isn’t frictionless, it’s productive friction. Resistance that teaches rather than resistance that blocks.
The second is design for the recurrent situations of the domain. Not for abstract task completion but for the specific, recurring patterns of breakdown and recovery that characterize how work actually happens in this particular context. What goes wrong here, regularly? What does recovery look like? What conversations does recovery require? Design for those situations, not for the idealized flow that precedes them.
The third is make commitments visible. If organizations run on networks of conversations, open requests, pending promises, unresolved assessments, then the most powerful thing a tool can do is make that network legible. Not by representing it as data but by supporting the speech acts that constitute it. By making it easier to make complete requests, to track open commitments, to surface the loops that aren’t closing before they generate breakdown.
This was actually the basis of Flores’ later software work, tools built around commitment tracking rather than task management. The difference sounds subtle. It isn’t. Task management treats work as things to be done. Commitment management treats work as promises made between people, with speakers, listeners, conditions of satisfaction, time frames, and histories of fulfillment that constitute trust over time.
The fourth principle, and the most philosophically grounded, is design for humans as they are, not as the model says they should be. Not for the rational actor who has clear preferences and consistent goals. For the thrown, backgrounded, breakdown-navigating being who discovers what they want partly through the process of trying to get it. Who changes their mind. Who brings concerns to a tool that the tool didn’t anticipate. Who is always already somewhere, carrying something, never arriving as a blank slate.
We are at the most consequential design moment in the history of computing.
AI systems are being embedded into workflows, decisions, creative processes, educational systems, medical diagnoses, legal reasoning, organizational coordination, at a speed that has no historical precedent. And almost none of it is being designed from the framework Winograd and Flores laid out in 1986.
It’s being designed from the rationalistic tradition. From the assumption that intelligence is processing, that knowledge is representation, that the relevant features of situations can be anticipated and formalized, that language conveys meaning reliably, that commitment is a feature to add rather than the substrate everything runs on.
The breakdowns are already happening. Confidently wrong outputs delivered without uncertainty. Commitment-shaped interactions with no commitment behind them. Tools that eliminate the productive friction that was building user competence. Organizations restructuring their conversation networks around systems that can receive speech acts but not make them.
Each one a predictable consequence of designing from the wrong model of mind.
Flores and Winograd weren’t writing a critique. They were writing a foundation. The foundation was largely ignored because the rationalistic tradition was producing impressive results and the commercial pressure was all in one direction.
The results are still impressive. The pressure is still in one direction. And the foundation is still largely ignored.
But breakdown is arriving. At scale. In ways that are surfacing the assumptions that were always there, invisible, running the show.
Which means we’re at exactly the moment Winograd and Flores were preparing us for. The moment when the present-at-hand snaps into visibility. When the tool stops disappearing and becomes an object of genuine inquiry.
The question is whether we have the framework to see what the breakdown is showing us.
You do now.
@ Lee Greene