Independent indices of what AI agents can do
Measuring how much of the world’s work AI agents can take over
Balam Labs builds open, preregistered indices of AI agents at work, task by task, and publishes every rating, score and line of code behind the numbers. The first is the Agent Delegability Index, covering 433 occupations and 146 countries.
| # | Major group | ADI, % of task time at L2–L3 | ADI | 95% interval |
|---|---|---|---|---|
| 1 of 9 | 1 · Managers | 81.8 | 78.1–85.1 | |
| 2 of 9 | 4 · Clerical support workers | 78.3 | 74.7–81.7 | |
| 3 of 9 | 2 · Professionals | 69.9 | 67.9–71.9 | |
| 4 of 9 | 3 · Technicians and associate professionals | 64.4 | 62.5–66.4 | |
| 5 of 9 | 5 · Service and sales workers | 39.6 | 34.8–44.6 | |
| 6 of 9 | 8 · Plant and machine operators and assemblers | 18.4 | 13.1–23.8 | |
| 7 of 9 | 6 · Skilled agricultural, forestry and fishery workers | 17.4 | 14.0–21.3 | |
| 8 of 9 | 7 · Craft and related trades workers | 17.4 | 15.5–19.5 | |
| 9 of 9 | 9 · Elementary occupations | 12.3 | 9.1–16.2 |
| # | Major group | ADI3, % of task time at L3 (scale 0–10) | ADI3 |
|---|---|---|---|
| 1 of 9 | 4 · Clerical support workers | 18.4 | |
| 2 of 9 | 3 · Technicians and associate professionals | 4.5 | |
| 3 of 9 | 5 · Service and sales workers | 3.4 | |
| 4 of 9 | 2 · Professionals | 2.4 | |
| 5 of 9 | 8 · Plant and machine operators and assemblers | 1.7 | |
| 6 of 9 | 1 · Managers | 1.5 | |
| 7 of 9 | 6 · Skilled agricultural, forestry and fishery workers | 0.7 | |
| 8 of 9 | 7 · Craft and related trades workers | 0.4 | |
| 9 of 9 | 9 · Elementary occupations | 0.2 |
| Major group | Physical presence | Accountability | Trust and care | Access to systems |
|---|---|---|---|---|
| 1 · Managers | 21% | 36% | 52% | 74% |
| 2 · Professionals | 33% | 45% | 50% | 64% |
| 3 · Technicians and associate professionals | 44% | 40% | 38% | 71% |
| 4 · Clerical support workers | 35% | 11% | 18% | 70% |
| 5 · Service and sales workers | 72% | 19% | 44% | 61% |
| 6 · Skilled agricultural, forestry and fishery workers | 90% | 27% | 17% | 62% |
| 7 · Craft and related trades workers | 88% | 22% | 8% | 44% |
| 8 · Plant and machine operators and assemblers | 86% | 30% | 14% | 66% |
| 9 · Elementary occupations | 94% | 16% | 11% | 45% |
Share of task weight flagged by each barrier. Barriers are not exclusive. Bold: 60% or more.
ISCO-08 major groups. Employment-weighted means over 433 unit groups; whiskers are 95% bootstrap intervals.
of task time in world employment is at least partly delegable to AI agents in 2026. Only 2.5% can be handed over in full.
Latest reports
Recent releases and what they found.
The Delegation Divide: ADI Report 2026.1
About a third of the world’s working time is at least partly delegable to AI agents, and full delegation is rare.
- 35.2% of task time in world employment can be at least partly delegated; only 2.5% in full.
- Delegable work rises with income, from 22.5% in low-income to 48.7% in high-income countries.
- Full delegation is most likely for tasks that take a person about 11.8 minutes.
ADI v2026.1 dataset on Zenodo
Every rating, score and line of code behind the index, under CC BY 4.0 and MIT.
- Ratings for all 18,796 O*NET task statements on four delegation levels and four barriers.
- Scores with 95% intervals for 433 ISCO-08 occupations and 146 countries, as CSV, Parquet and a static JSON API.
- A Python package and an MCP server that return each record with its citation. doi:10.5281/zenodo.23552918
ADI preregistration on OSF
Hypotheses, rubric and analysis plan were registered before the main scoring run.
- Four hypotheses: the income gradient, the gap with observed AI use, agreement with earlier exposure indices, and task length.
- Deviations from the registered plan are dated and listed in the methodology.
- The registration is under embargo until 31 January 2027.
Products
Each index is versioned, has a DOI and keeps its numbers at fixed paths.
Every record is citable and machine-readable: data, API, Python package and MCP server.