In the preceding essay I put the competence problem at the moment of appointment; in this one I ask who carries the cost of it, and the answer sits in Champniss's own data, largely unremarked.
Employees in their first five years of working life record a psychological debt score related to their use of AI of fifty-four. For those past their twentieth year this debt records forty. The explanation offered in the HBR article is that the junior employees feel a greater need to demonstrate technical knowledge, which the machine threatens, while senior people have already demonstrated leadership capabilities lying beyond its scope. Plausible enough, and flattering to those with senior in their title or their hair colour. There is however a plainer reading available, and it requires no psychology whatever. The burden falls heaviest on those least able to push back against it.
Consider the two positions concretely. A graduate analyst instructed to put her first draft through the model has no standing to refuse and no capital to spend on refusing; her competence is unestablished, her tenure is short, and a reputation for being difficult about tools is not one she can currently afford. An executive at the other end of the career spectrum is differently placed entirely. They can decline without consequence, and a striking number do—continuing to have their doing work done by people rather than machines, while requiring use of the machine by everyone below them. The mandate travels downward. The typical command and control structure too many organisations adopt.
The Drinks Cabinet
There is an older test for this, and it does not need a survey instrument. If a rule binds the crew and not the bridge, the rule is not about the conduct. It is about position. An office in which the manager may drink at their desk and the coordinator may not is not operating a policy on alcohol; it is operating a policy on rank, and everyone in the building understands this by the end of their first week with the organisation. The specific liberty is beside the point. What is being communicated is that the rules exist to protect those who make them, which is corruption in the structural sense—not theft, not fraud, but the quiet bending of a system away from its purpose and toward the comfort of its incumbents.
The organisational research on this is unambiguous, and it is not primarily about outcomes. People assess the fairness of a workplace less by what they receive than by whether the process that allocated it applied consistently and without bias. Most tellingly of all, procedural fairness of that kind predicts commitment, trust, and corporate citizenship behaviour more powerfully than the distribution of benefits itself. Robert Jackall's field study of managerial life put the mechanism plainly forty years ago: what is right in the corporation is what the person above you wants, and moral reasoning inside such a hierarchy tends to organise itself around the protection of position rather than the pursuit of purpose. An adoption mandate that exempts the people who handed down the mandate is not an anomaly within that system. It is the system behaving normally.
So when Champniss records that employees fear a loss of credibility from being seen to use the tool, and proposes bringing use into the open as the remedy, the diagnosis has stopped one floor short. The fear experienced is not a psychological condition. It is an accurate reading of the organisation's actual reward structure by people who are reading it accurately. Where visible reliance on a machine is understood to diminish authorship, and where the standing that survives such diminution is concentrated at the top, concealment below is simply rational. The remedy is not a culture campaign encouraging candour about tool use. It is governance willing to establish that the rule binds the bridge, which is a decision no communications programme can make on its behalf.
The Perspex Award
The AI mandate requires a measure, and the measure too often used is that of consumption. Brett Raven, a fractional CIO who is called in when a transformation has stalled, described to me on this programme what he has watched organisations do with it: pumping token usage upward because the number looks like progress, and in some cases using token spend to measure an individual's productivity—which, as he put it, translates directly into currency, so that a person who is not spending enough money in their job is deemed not to be doing it properly. His image for what this earns is the perspex award from the model vendor, the little acrylic block on the desk thanking the recipient for spending ten million tokens this month. My own question to him was what such a thing unlocks: ten thousand tokens spent, a plastic trophy acquired, and no answer to what was actually gained. The metric he distrusts most, he said, is hours saved—not because time is never saved, but because he would like to see where the hours went before celebrating the success.
This is proxy substitution reaching its terminal form. A score at least gestures at the condition it claims to measure. Token consumption does not measure an outcome at all; it measures an input, and specifically it measures a cost. An organisation optimising for it is rewarding expenditure and calling the result adoption. That such a metric could establish itself in serious companies tells us something about the appetite for legibility that this series has been describing from the first essay: when judging value is hard and counting spend is easy, the count wins, and it wins fastest in the organisations least equipped to argue with it.
