Should you get less credit because AI made it easy?
An AI company-design proposal runs into a human problem: recognising valuable new work without turning everyone's past contribution into a live scoreboard.
A stable foundation, with bounded commitments above it.AI-generated conceptual setting.
Imagine two people solving the same expensive problem. One takes a month. The other uses AI and finishes in an afternoon.
Paying by time seems to penalise the second person for finding a better way. Paying only for the result raises another question: whose earlier research, relationships and failed attempts made that afternoon possible?
That tension sits inside an AI company-design proposal I have been exploring. It tries to recognise future contributions without making everybody’s existing position subject to continuous recalculation.
The interesting part is what the proposal refuses to combine. The past agreement and the next piece of work do not have to be settled by the same score.
Follow the idea
What happens when a month of work takes an afternoon?
A company-design proposal about existing commitments and future work. These shapes are not ownership percentages, legal instruments or an adopted agreement.
The foundation stays in view at every step.
The earlier agreement stays visible
A new tool does not erase the conditions under which people previously contributed. Existing rights are not silently recalculated by an activity score.
The foundation stays in view at every step.
Agree on the next question
A finite offer states the work and acceptable evidence before the result. A useful investigation may conclude that the feature should not be built.
The foundation stays in view at every step.
Decide what the evidence earns
The evidence can show what happened without automatically settling the reward. People retain a way to question the result and the decision.
The past is not another task to rescore
An earlier contributor may have taken a risk when the project was uncertain. A later contributor may produce something more quickly with tools that did not exist then. Both facts can be true.
A system that repeatedly asks who is contributing most now can erase the circumstances under which earlier work happened. It can also turn every new task into a negotiation about everybody else.
The proposal begins with existing rights and commitments, then defines a finite offer for future work. What is being asked? What would count as doing it? What is the proposed reward? Those questions are addressed before the result arrives.
This does not make every existing arrangement fair, or prevent people from renegotiating. It keeps an activity-scoring system from quietly doing the renegotiation for them.
Would the company pay for an answer it dislikes?
Suppose a company wants an automation built. Before committing to it, someone is asked to test whether it can handle an important exception.
The investigation finds that it cannot.
If the reward depends on delivering the feature, the investigator has a reason to soften the finding or keep building. If the agreed task was to resolve that uncertainty, a careful negative result can be exactly the contribution the company needed.
That is a more demanding idea than paying for anything labelled a learning. The question and acceptable evidence need to be clear before the answer is known. Otherwise an unsuccessful attempt can be given a flattering explanation afterwards.
It also leaves a difficult judgement: was the investigation good enough? AI can help assemble evidence and expose gaps. A tidy report cannot settle the fairness of the reward by itself.
People are more than the work the system can see
A contribution system has a bias toward things it can count. Documents, commits and completed tasks leave traces. Preventing a bad decision in a conversation may leave very little.
Asking everyone to produce more traces can change the work. People may become better at demonstrating contribution while the company becomes worse at recognising it.
The proposal therefore uses bounded pieces of work rather than a permanent score for each person. It still needs ways to contest a decision, recognise work outside a neat task and stop collecting evidence when the question has been answered.
Existing shareholder rights and company share issues also have legal requirements that a software record cannot replace. This is a design exploration, not an adopted agreement or advice about percentages, instruments or tax.
The proposal’s own attempted simulation ran into a revealing problem, discussed in the companion article. It could not establish which reward method was better. Designing a transparent process is already difficult; claiming to have calculated fairness requires much more.