Have you felt a subtle friction in your team lately — a disconnect between how people want to build and how they are actually being evaluated? You are not alone, and it is not a culture problem in the way it is usually diagnosed.
We have integrated extraordinary AI tools into our workflows, but we are still measuring human value with performance systems built for a completely different era. The tools changed in eighteen months. The performance framework has not changed in twenty years, and it was designed to optimise something we no longer want.
The tension, stated plainly
Traditional performance models reward three things: predictable execution, adherence to a strict job description, and the absence of mistakes. Every element of the standard apparatus — annual objectives, calibration sessions, ratings distributions — is tuned to detect and reward exactly those.
AI-native talent operates on a different frequency entirely. They are not primarily completing tasks. They are multipliers — orchestrating ecosystems of AI agents, tools and ideas, and producing leverage rather than output. Their contribution is often invisible to a system that counts deliverables, because their best week may have produced no deliverable at all and instead removed the need for a category of work.
The collision is easy to see once you name it. This kind of person runs ten bold experiments and fails at eight of them. Under a traditional model, that is an eighty percent failure rate and a difficult conversation in the calibration meeting. Under any sensible reading of what actually happened, it is two discoveries that would not otherwise exist, purchased cheaply.
What the punishment actually costs
When we penalise those eight failures, we do not simply slow productivity down. We do something more expensive and much harder to reverse: we exhaust their spirit. We tell a curious person, in the only language an organisation speaks fluently, that their curiosity is a liability.
People do not usually argue with that message. They adapt to it. They run two safe experiments instead of ten. They stop volunteering the idea that might not work. They become, within a couple of cycles, exactly the predictable executor the system was measuring for — and the organisation concludes, wrongly, that it never had innovative talent in the first place.
This is the quiet version of the problem. There is no incident, no attrition spike, nothing that shows up on a dashboard. Just a slow narrowing of what people are willing to try.
Kimi’s philosophy, and why it stayed with me
The evaluation and performance philosophy published by Kimi is the clearest articulation I have seen of the alternative, and it is worth studying not for its specific mechanics but for what it chooses to treat as signal.
The move is to stop evaluating compliance with a defined role and start evaluating the size of the leverage a person creates — including the leverage created by the experiments that failed. Failure stops being a deduction and becomes information the organisation paid for and now owns. That single reframe changes the incentive at the exact point where most performance systems break.
Four questions worth asking about your own system
- What does a great year look like for someone whose main contribution is orchestration? If your framework cannot describe this without stretching, it cannot rate it fairly either.
- Where do failed experiments go? If the honest answer is “into the rating,” you have priced curiosity and the price is high.
- Is the job description a floor or a ceiling? Strict role definitions were a fairness mechanism when work was stable. In an environment where the useful work is being redefined continuously, the same mechanism becomes a cage — and the best people notice first.
- Who is actually rewarded for reinventing how work gets done? Microsoft’s 2026 Work Trend Index put this at 13% of people. If you have not deliberately designed against that number, assume you are inside it.
Bridging the gap
None of this argues for abandoning accountability, and I want to be careful about that, because the loose version of this argument does real damage. Predictability genuinely matters in the parts of an organisation where it matters — safety, compliance, financial control, anything regulated. The error is applying a single ruler across the whole organisation and then being puzzled that the people building the future keep coming up short on it.
The bridge is not a softer system. It is a more honest one: different work, measured for what it actually produces. Execution work measured on reliable delivery. Multiplier work measured on leverage created, learning generated, and capability left behind — with failed experiments counted as the cost of information rather than as evidence of a weak performer.
What does your organisation reward — predictable execution, or the leverage that changes what execution means?
Whatever the answer is, your best people already know it. They worked it out from the ratings, not from the values statement.
📚 Start here: AI-Native Organization Design — the full research hub, with every article and video in one place.
More in this series
- 67% of AI Adoption Blockers Are Organizational, Not Individual
- $2M Revenue Per Employee: The Case for the Living Organization
- The Living Organization: Five Pillars of an AI-Native Operating Model
- Why Hierarchy Existed — and Why AI Just Removed the Reason
- Is HR Ready to Design Human and AI Agent Productivity Together?
- How Much Does a CEO Really Need to Understand About AI?
- The Five Steps of AI Adoption: From Gated to AI-Native
- Your AI Isn’t Failing — Your Specification Is
Watch the full series
This article accompanies a video from AI-Native Operating Model and Org Design — a series on how organisations actually absorb AI, and where they break.
- ▶️ Watch the full playlist — AI-Native Operating Model and Org Design
- 📺 Subscribe on YouTube — @BreezeDONG
Chunfeng “Breeze” Dong is an executive coach (ICF PCC, CPCC) and founder of Springbreeze Ventures, with twenty years in organisational development inside Fortune 100 companies — Roland Berger, Siemens, ABB and Roche. She writes on AI-native organisation design, human–agent governance and change leadership.
📘 The Living Organization · 📘 A Soulful Transition · 🔗 LinkedIn
