Start with tasks
We describe recurring work, frequency, responsible owner, oversight, and the human moat.
How the replacement index works
A deterministic editorial model estimates task coverage and implementation practicality. It does not predict layoffs, individual performance, or the future of a profession.
Score capabilities + constraints
Ten 0–100 subscores feed a documented weighted model.
Apply safety overrides
High risk, physical dependency, or mandatory sign-off prevents an “Automate now” verdict.
Review editorially
Every role has notes, a date, confidence, and a cautious plain-language conclusion.
Weights
The deterministic scoring helper
Positive weights increase automation practicality; negative weights reflect the human and risk burden.
| Signal | Weight | Editorial question |
|---|---|---|
| Digital task share | 17% | How much work happens in software and structured information. |
| Task repeatability | 13% | How consistent the inputs, rules, and outputs are. |
| Model / tool capability | 14% | How reliably current tools perform the task family. |
| Integration availability | 9% | Whether the necessary systems expose safe, practical interfaces. |
| Reliability & verifiability | 12% | Whether output can be checked against sources or deterministic rules. |
| Human judgment | −10% | Novel tradeoffs, tacit context, and accountable decisions. |
| Relationship / empathy | −8% | Trust, care, negotiation, and interpersonal nuance. |
| Physical dependency | −7% | Work that requires embodied presence or physical manipulation. |
| Consequence / regulatory risk | −6% | Potential harm and need for licensed or formal sign-off. |
| Supervision burden | −5% | Human effort required to review and recover from errors. |
Most standardized work can be automated with practical controls.
AI can own repeatable work; a human remains accountable.
AI assists strongly, but human judgment remains central.
Current AI is not a responsible substitute for this role.
Useful estimates, not scientific certainty
- Roles vary by company, jurisdiction, seniority, and tool access.
- Coverage describes recurring tasks that AI may address—not headcount reduction.
- Scores are reviewed when capability, regulation, or strong evidence changes.
- Sponsors cannot buy a verdict, a score, a rank, or favorable wording.
- Named employees and personal performance data are outside the product’s scope.