The Rising Floor, The Distant Ceiling

In the AI Story

The temptation the rising floor creates is mathematically coherent within a narrow frame. If AI handles 95% of situations adequately and expert judgment is needed only for the remaining 5%, the expected value of investing in expert development might appear to fall below the expected value of tool proficiency. The reasoning is seductive because the arithmetic is not wrong. It is dangerous because it ignores the nature of the 5% where expertise matters: those situations are high-stakes, high-consequence, and often irreversible. The expected value of expertise is not its frequency of deployment multiplied by its average benefit; it is the frequency multiplied by the magnitude of its benefit in the situations where it matters, and in those situations the magnitude is often enormous.

The medical domain illustrates this starkly. AI diagnostic systems now match or exceed experienced physicians across routine presentations. The radiologist's AI catches cancers. The dermatology AI classifies lesions. The clinical decision-support system suggests diagnoses. The floor has risen. But the cases that are not routine — atypical presentations of common diseases, rare diseases mimicking common ones, patients whose symptom constellations match no pattern in the training data, complications requiring the surgeon to abandon the planned approach and improvise — these are the cases where patients live or die based on the depth of the practitioner's mental representations. The 6-percentage-point drop in endoscopist polyp detection documented by Hosanagar, translated to screening population scale, is thousands of missed cancers. The floor rose. The capability to stand without it had not developed.

The software-engineering analog is structurally identical though less visceral. AI tools make competent code accessible to novices. But the systems that matter — systems handling millions of users, processing financial transactions, controlling medical devices, managing critical infrastructure — are systems where the floor is irrelevant and the ceiling is everything. The senior engineer's capacity to design systems that scale, to anticipate failure modes that testing cannot reveal, to make architectural decisions that determine whether the system survives its success — these capacities are built through thousands of encounters with systems that failed, through debugging production incidents under pressure, through accumulated understanding of how distributed systems behave under load. This understanding cannot be shortcut. Only a practitioner who has built the representations through deliberate engagement with the specific challenges of large-scale systems can provide it.

The organizational consequence the framework specifies is that assessment must separate output quality from practitioner capability. These were reasonable proxies before AI; they no longer are. New methods are needed — code reviews without AI assistance, diagnostic exercises without decision support, design challenges requiring first-principles work — that evaluate understanding rather than production. Organizations that continue treating output quality as capability indicator will systematically overestimate their workforce's ability to handle the situations where the ceiling matters. The error will remain invisible until the situation arrives, and in the situations where the ceiling matters, invisibility until the moment of revelation is precisely the worst possible failure mode.

Origin

The rising-floor observation is widely made across the contemporary AI discourse; the specific analytical frame deployed here combines it with Ericsson's expertise research to specify what the shadow consequence is and why it matters. The argument draws on MIT Sloan Management Review's 2025 work on 'meta-expertise' and on the deskilling literature documenting tool-free capability erosion.

Key Ideas

Floor and ceiling are independent. Tool-enabled rising floor does not raise the expertise ceiling; the two respond to different mechanisms.

Expected value arithmetic is misleading. Expertise's value is concentrated in rare, high-stakes situations whose impact per deployment is not averageable across all deployments.

Assessment must be redesigned. Tool-assisted output is no longer a valid proxy for tool-independent capability.

Investment pressure runs against maintenance. Organizational incentives favor proven-rising-floor visible wins over invisible ceiling-maintenance.

Invisibility until revelation. The capability deficit manifests exactly when the capability is most needed, making the miscalibration catastrophic precisely when it cannot be corrected.