Does Value Optimisation Serve People or Extract from Them?
The instalment that argues with this site most directly.
This instalment was first published under the site’s early, informal name for connecting every unit of AI work to a measured outcome, “Valuemaxxing”. The site now uses plainer terms for the same discipline: AI value attribution and AI value management. The question below applies to both.
The Question
Value optimisation says connect every AI unit to a measured outcome and optimise for value. But value to whom? Is this honest stewardship or an extraction engine with a dashboard?
The optimist's case
Measurement is not the enemy of the workforce. Unmeasured AI is.
The alternative to measuring is not a humane absence. It is tokenmaxxing, vanity dashboards, and hype. The people hurt by those were rarely shareholders.
Evidence Token leaderboards put surveillance pressure on employees. Agents on unredesigned workflows made jobs worse. Spending cuts came from headcount, not the model bill.
Undisciplined AI already extracts. Measurement catches it. The discipline kills shelfware, penalises retries and failure through cost per successful outcome, and demands a counterfactual before any productivity claim.
The four value dimensions include risk reduced. Interpretation Counting risk reduction as value is structurally on the side of people who bear the risk.
Summary: measurement is not the enemy of the workforce. Unmeasured AI is. Visibility is the precondition for fairness.
The sceptic's case
The framework optimises whatever you point it at. History says who does the pointing.
The sceptic grants all that, then asks: who defines the outcome?
Every metric is agnostic about whose value it serves. Point it at shareholder return and it faithfully optimises extraction. Fewer people, more monitored work, surplus captured upward. The dashboard is impeccable. That is the problem.
Interpretation The structural worry: what is easy to measure crowds out what matters. Labour cost removed is legible this quarter. Institutional knowledge lost, skills atrophied, trust burned, work intensified: all real, all slow, all invisible to the optimiser.
An optimiser fed only the legible trades the invisible for it. Well-run value optimisation is a very good optimiser.
Evidence The precedent is a century of scientific management. The gap between value created and wages paid in heavily measured work is not hypothetical.
Forward bet: within a few years, a value dashboard will justify a workforce reduction whose costs were not on the dashboard.
Summary: the framework optimises whatever you point it at. History says who does the pointing.
Synthesis · house view
Watch whether value shows up across all four dimensions or collapses to labour cost removed
This will not be settled by a single study. Watch honestly.
Track deployments: does realised value show up across all four dimensions, or does productivity gained collapse to labour cost removed?
Do organisations redeploy or reskill the time saved, or do they harvest it? Does anyone put worker-facing measures on the value dashboard? Do those measures move?
Interpretation A discipline that serves people will show it in its own metrics.
If We Get It Right / If We Get It Wrong
Right + honestly governed: AI value is real, shared, and demonstrable. The framework serves all stakeholders.
Right + captured: The machinery works, but only for owners. Value extraction is efficient and well-measured.
Wrong + abandoned: Back to hype and tokenmaxxing. The alternative was not better.
Wrong + unexamined: Measurement regimes intensify work without improving outcomes. The dashboard becomes a tool of extraction.
The Author’s Honest Position
The sceptic is right about the core point, and the framework should say so plainly. The definition of value is a governance choice, not a technical one. No metric makes that choice for you.
The machinery is neutral. Neutral machinery is available for extraction. The defence is not in the metrics. It is in the base of the iceberg.
Position: run value optimisation with the value definition set openly. Put people-facing measures on the same dashboard as financial ones. Make the workforce able to see what is being optimised in their name.
Interpretation The honest tell: if everything on the dashboard faces the shareholder, believe what you are looking at.
I think the discipline, openly governed, serves people better than the alternatives. But that is a bet about governance, not mathematics.
I hold this with less certainty than anything else in the series.