Designing for the Bounded Builder
Designing for the bounded builder is the practical prescription that Simon's framework generates for the AI age. The premise: AI has expanded what individual builders can generate but not what they can evaluate. The binding constraint has shifted to judgment, attention, and the integrative capacities that bounded rationality imposes strict limits on. The design response must therefore shift too — away from structures that optimize generation (the old hierarchies, the old specializations, the old productivity metrics) and toward structures that support evaluation (new decompositions, new pedagogies, new personal practices). The prescription operates at three scales: organizational (building attention architectures that protect evaluative capacity from the flood of AI-generated output), educational (teaching evaluation rather than merely generation, building pattern libraries that enable builders to assess what they have produced), and personal (establishing disciplines that impose representation investment before search, evaluation checkpoints during production, and reflective pauses that prevent the satisficing threshold from rising past the point of diminishing returns). Each scale requires its own design, but all share the same underlying principle: treat the builder's bounded attention as the scarce resource, and allocate it with the deliberateness that well-designed systems have always allocated their scarce resources.
In the AI Story
The organizational scale requires rebuilding the attention architecture that the old hierarchy provided. The traditional corporate hierarchy was a cascading attention filter — each level condensing information for the level above it. AI has bypassed the filtering by enabling individuals to produce high-quality output that flows directly to decision-makers whose attention was previously protected. The response is not to restore the old hierarchy but to build new filtering mechanisms — vector pods, evaluation gates, protected reflection time — that serve the filtering function in the new cognitive environment.
The educational scale requires shifting pedagogical investment from generation to evaluation. The classical curriculum trains builders to produce: to write code, to draft briefs, to build models. AI has made production abundant. What remains scarce is the evaluative capability that distinguishes good productions from merely functional ones — the pattern libraries, the meta-cognitive skills, the domain expertise that determine whether a bounded mind can assess AI-generated output wisely or only superficially. Building this capability requires pedagogy focused on assessment rather than creation, on judgment rather than execution, on the kind of slow experiential accumulation that deliberate practice requires.
The personal scale requires individual disciplines that impose structure on the AI interaction. The default interaction rewards speed, volume, and seductive acceptance of confident output. The disciplined interaction — setting explicit goals before engaging the AI, imposing evaluation checkpoints, maintaining awareness of the tool's choice architecture, requesting alternatives that the tool did not volunteer — works against the default dynamics and is therefore cognitively costly. The cost is the price of evaluative wisdom in an environment that rewards evaluative speed.
Origin
The prescription emerges from Simon's framework applied to contemporary conditions. Simon himself advocated for the design of institutions for bounded agents throughout his career, insisting that the science of the artificial was a discipline as rigorous as any natural science and more urgent for practical outcomes. The AI age has made his prescription impossible to ignore, though whether the response will be adequate to the challenge remains an open question.
The specific prescriptions at each scale are being developed in real time by researchers, practitioners, and institutions responding to the empirical reality of AI-augmented work. The Berkeley study's AI Practice framework represents one version of the organizational prescription. Emerging curriculum reforms at universities that take AI seriously represent early versions of the educational prescription. Individual practitioners experimenting with structured prompting, evaluation disciplines, and reflective pauses represent the personal prescription. None of these is yet mature; all of them are in the phase of productive experimentation that design knowledge requires before it can stabilize into durable practice.
Key Ideas
Generation-optimization is obsolete. Structures designed to maximize what builders produce fail in an environment where production is abundant and evaluation is scarce.
Three scales require redesign. Organizational, educational, and personal structures all need to be rebuilt around the bound that remains: bounded evaluative attention.
Evaluation architecture is the new organizational challenge. The filtering function that hierarchy performed must be performed by new structures adapted to the AI environment.
Pedagogy must shift to judgment. Teaching evaluation requires different methods than teaching generation — the pattern libraries and meta-cognitive capacities built through experiential practice.
Personal discipline counteracts defaults. The structured interaction practices that produce wise AI use work against the default dynamics and must be deliberately cultivated.
Appears in the Orange Pill Cycle
Further reading
- Simon, The Sciences of the Artificial (1969)
- Simon, Administrative Behavior (1947)
- Cal Newport, Deep Work (2016)
- Donald Schön, The Reflective Practitioner (1983)