Skip to Content
DocumentationBriefsMGI: Agents, Robots, and Us

Agents, Robots, and Us

A cheatsheet on MGI’s methodology for assessing skill partnerships in the age of AI — covering automation potential, adoption modeling, occupation archetypes, and the Skill Change Index.

Source: MGI Technical Appendix  — November 2025


Scope of the Analysis

The research focuses exclusively on paid productive hours in the US workforce, spanning full-time and part-time work across industries, functions, occupations, and skill levels. Only time awake spent on work-related activities is assessed — roughly 45 percent of waking hours. Unpaid tasks and leisure are excluded, though agents and robots could support productivity in those areas as well.

The methodology was first developed in 2017 and refreshed for this latest research, drawing on the BLS O*NET breakdown of ~800 occupations into ~2,000 detailed work activities.


The Automation Model

MGI’s approach compares the capability level required to perform each detailed work activity (DWA) against 18 human capabilities with automation potential. An updated expert survey of AI researchers provides current performance estimates and future trajectories across those capabilities.

Employment and wage data are refreshed at the BLS occupation level. Together, these inputs produce an estimate of technical automation potential in 2025 — the frontier of what is currently demonstrable.


Adoption Model: Three Steps

StepDescription
1. Solution timelineEstimate the time to implement a solution that can automate each DWA once all capability requirements are met by technology.
2. Cost parityEstimate technology costs at introduction, declining over time per historical precedent. Adoption begins when automation costs reach parity with human labor.
3. Diffusion curvesModel timelines from initial uptake to plateau using S-shaped (sigmoidal) curves. Plateau adoption historically takes one to three decades, capturing regulation, investment, management decisions, and user preferences.

Activity Classification

Each of the ~2,000 DWAs is classified by automation potential and the capabilities it requires:

  • Activities that are not yet technically automatable remain people-performed
  • Among automatable activities, those requiring physical capability (gross motor, fine motor, mobility) are classified as robot-performed
  • Those relying only on cognitive or social-emotional capabilities are classified as agent-performed

Seven Occupation Archetypes

Occupations are grouped into archetypes based on the share of work hours devoted to people, agent, or robot activities. The threshold for “centric” classification is >55% of hours in a single category.

Core Archetypes:

  • People-centric — e.g. Registered nurses (~70% people hours)
  • Agent-centric — Majority cognitive/social-emotional tasks
  • Robot-centric — Majority physical capability tasks

Combined Archetypes: Four additional archetypes cover roles with balanced mixes: people-agent, people-robot, agent-robot, and near-even splits across all three. Combined roles spend >55% of hours across two activity types.


Skill Change Methodology

The skills analysis integrates four inputs:

  1. Occupation-level employment data
  2. DWAs per occupation
  3. Skills relevant to each DWA
  4. The automation adoption model

Data sources include BLS (~800 occupations), O*NET (~2,000 DWAs), and Lightcast (~34,000 skills across ~2,000 occupations).

Skills appearing in fewer than 5% of job postings per Lightcast occupation were filtered out, narrowing to ~7,000 skills. GPT-4o was used to create ~3.4 million occupation-DWA-skill mappings, validated against a manually-built 1,000-cell template with iterative quality testing.


Skill Classification

TypeDefinitionExample
People-led>55% of time in non-automatable activitiesConflict resolution
AI-led>55% of time in automatable activities (agent-led for cognitive, robot-led for physical)Machine operation
SharedBetween thresholds — requires combination of people and AIDetail orientation

The Skill Change Index (SCI) is calculated from the midpoint automation-adoption rate projected for 2030 across occupation-DWA combinations mapped to each skill, weighted by time spent.


Key Data Points

CategoryScale
Occupations~800
DWAs~2,000
Skills~7,000
Occupation-DWA-skill links~3.4M

Sources: BLS/O*NET for occupations and activities, Lightcast for skills, Census CPS for employment, AI expert survey for capability assessment


The Bottom Line

These archetypes and scores reflect technical automation potential, not predictions. Actual adoption depends on solution timelines, cost dynamics, and diffusion speed. The Skill Change Index provides a single, comparable measure of how sensitive each skill is to automation — making it possible to prioritize workforce planning and reskilling investments.

Last updated on