AI-Wiki

Agents, robots, and us: Skill partnerships in the age of AI

21 min read

Yee / MGI — Agents, Robots, and Us: Skill Partnerships in the Age of AI

The McKinsey Global Institute’s November 2025 flagship report on the workforce implications of AI agents and robots — the wiki’s most structural and panoramic anchor to date on the question what will people, agents, and robots do in 2030, and what skills will that require? Authored by seven McKinsey/MGI partners and senior fellows under MGI director Lareina Yee (Bay Area office) with MGI chairman Sven Smit (Amsterdam) as co-author, the report combines BLS + O*NET + Lightcast labour-market data with MGI’s automation-adoption model and a new Skill Change Index (SCI) built using OpenAI’s GPT-4o for skill→DWA mapping at scale (~3.4 million occupation-DWA-skill links). MGI academic advisers: Nobel laureate Sir Christopher Pissarides (LSE) and Matthew J. Slaughter (Tuck/Dartmouth), with Luca Vendraminelli (Stanford Digital Economy Lab / Stanford HAI) as postdoctoral research contributor.

The wiki’s first MGI workforce-and-skills layer panorama, complementing the March 2026 MGI Race-Takes-Off industry-and-competition layer panorama from the same intellectual project. Together the two reports constitute MGI’s two-layer panorama of the AI-economy question: where will the value migrate (Race Takes Off, 18 arenas) × who and what will do the work that captures it (Agents, Robots, and Us). The MGI Race Takes Off ingest’s open follow-up — update McKinsey & Company with MGI sub-section — is fulfilled by this ingest’s promotion of McKinsey Global Institute to its own entity page (3rd-mention rule).

TL;DR

  • The headline framing: AI will reshape the building blocks of work — the skills — not merely automate jobs. “Work in the future will be a partnership between people, agents, and robots — all powered by AI.” The report uses “agents” and “robots” as broad practical terms for nonphysical and physical automation respectively, not narrow AI-paper definitions.

  • Technical automation potential: ~57% of current US work hours. Today’s demonstrated technologies could in theory automate just over half. Not a forecast of job losses — a measure of the technical frontier. Adoption will take decades (electricity: 30+ years; industrial robotics: similar; cloud: 2/3 of corporate workloads still off-cloud as of 2023 “despite the technology being widely available since the mid-2000s”).

  • The agents/robots split: 65% of US work hours require only nonphysical capabilities (agent territory), 35% require physical capabilities (robot territory). Of the 57% automatable: 44% via agents + 13% via robots. The remaining 43% remains people-only (21% nonphysical-only + 22% physical-only requiring social/emotional capabilities).

  • The Skill Change Index (SCI) — a time-weighted measure of automation’s potential impact on each of ~6,800 employer-cited skills. Computed by mapping ~7,000 filtered skills (>5% of postings) to ~2,000 O*NET DWAs across ~1,800 Lightcast occupations using OpenAI GPT-4o for skill→DWA labelling (~3.4M mappings; manual 1,000-cell template for validation; iterative prompt refinement). Skills are classified people-led / shared / AI-led (further: agent-led / robot-led) at a 55% threshold.

  • The 72% shared-skills finding — most human skills will remain relevant, but how they’re used will evolve. ~72% of today’s skills are required for both automatable and non-automatable work; 11% are required for people-led work only; 17% are required for AI-led work only.

  • AI fluency demand has grown sevenfold (6.8×) in two years (2023→2025: 1.0M → 7.0M US employees in occupations where job postings call for AI-fluency skills). Technical AI skills (Govern AI + Develop AI; 55 skills) grew 1.6× (2.1M → 3.3M). Total AI-related skills (66 skills): 2.2M → 7.5M (+3.5×). AI fluency is rising faster than any other skill in US job postings.

  • The seven occupation archetypes (Exhibit 3, ~800 BLS occupations classified by share of work hours; “centric” = 55%+ in one activity type):

    ArchetypeWorkforce shareAvg payExample occupations
    People-centric34%$71kRegistered nurses, psychologists, firefighters
    People-agent21%$74kSales reps, secondary school teachers, HR specialists
    Agent-centric30%$70kAccountants, software developers, lawyers
    People-robot<1%$54kInsulation workers, drywall + ceiling-tile installers
    Robot-centric8%$42kStockers + order fillers, welders, cooks
    People-agent-robot5%$60kReceptionists, medical assistants, correctional officers
    Agent-robot2%$49kMachine setters, bakers, library assistants

    “People-centric” (34%) and “agent-centric” (30%) dominate — most US workers are at the polar ends of the human-vs-AI cut. The agent-centric quadrant is where the wage premium is at risk: $70k average pay, mostly knowledge-work occupations (accountants, software developers, lawyers) — “highly automatable jobs… large shares of cognitive tasks that could technically be handled by AI systems”. The people-centric end ($71k average pay, healthcare + safety + interpersonal) is largely insulated by physical activity that current technology cannot replicate.

  • The eight high-prevalence skills that endure across the workforce (Exhibit 11):

    SkillRelevance %Agents/Robots willPeople will
    Communication99Generate content, accelerate data flowRefine nuance and storytelling
    Management94Automate scheduling and monitor metricsCoach and lead hybrid teams
    Operations84Execute routine tasks, optimise efficiencyDesign smarter processes and strategise
    Problem-solving83Identify patterns and propose optionsInterpret findings and make judgments
    Leadership83Drive change, support decision-makingGuide and motivate teams
    Detail orientation80Run quality checks, flag anomaliesAudit outputs and validate outcomes
    Customer relations80Route requests, handle routine queriesStrengthen loyalty and build relationships
    Writing76Produce drafts, propose revisionsRefine text and craft story

    The framing — “AI does the quantity, humans do the quality” — is convergent with Storoni’s neuroscience-vantage reframe of efficiency for the AI era (gear-3 quantity vs gear-2 quality), now at MGI structural-modeling altitude.

  • The SCI’s three skill-evolution paths:

    1. Highly exposed (top quartile) — specialised skills like specific programming languages, accounting processes → demand will decline.
    2. Middle quartiles — transferable skills + AI fluency → demand will evolve in nature and application rather than rise or fall.
    3. Low exposure (bottom quartile) — grounded in human connection and care, leadership, healthcare skills → will endure.

    By skill category (Exhibit 13): digital skills 42% high-exposure, information skills 29% high, machinery 27% high, construction 31% high, communication-creativity 14% high, management 22% high, handling+moving 29% high, assisting+caring 10% high. Assisting + caring is the most-protected category; digital skills are the most-exposed.

  • AI-skill demand concentration: ~75% of US AI-skill demand comes from 3 occupation groups (Computer & mathematical: 2.6M, Management: 2.3M, Business & financial operations: 0.8M). 9 occupation groups (~40% of workforce, lower-income roles incl. construction, transportation, production, food service, healthcare support) have near-zero AI-skill demand in postings — the AI-skill labour market is currently bifurcated.

  • Demand for “AI-adjacent” capabilities is rising — process optimisation (+114 occupations 2023-25), quality assurance (+87), business analysis (+90), teaching (+90), business intelligence (+76), people management (+138), software development (+65). Demand for skills AI already performs well is falling: routine writing (-134), mathematics (-133), basic technical knowledge (-115), general science and research (-140), customer service (-83), general accounting (-70), office productivity tech (-69), billing and invoicing (-49).

  • The $2.9 trillion US economic value claim — under the midpoint adoption scenario, AI-powered agents and robots could generate ~$2.9T/year US economic value by 2030. Globally: $28.7T (up from earlier ~$26T MGI projection). Splits:

    • 77% from agents = $2.26T; 23% from robots = $0.67T. “Agents could contribute more than three-quarters of the economic value of AI and automation.”
    • By concentration: ~60% of gains concentrated in sector-specific workflow domains ($1.7T: Knowledge $773B + Frontline $424B + Production $556B); ~40% in cross-cutting domains ($1.2T: IT $196B, Sales $90B, Marketing $167B, Legal $21B, Risk/compliance $11B, Logistics $122B, Customer support $60B, etc.).
    • By sector (top 5): Administrative+support+government $368B, Healthcare+social assistance $351B, Professional+scientific+technical $323B, Educational services $256B, Manufacturing $250B.
    • By sector adoption rate (avg automation as share of current work hours, midpoint scenario): 31% manufacturing, 30% finance/insurance + accommodation/food, 29% utilities + mining + agriculture, 28% information + management of companies, 27% wholesale + other services, 26% real estate + arts + retail + administrative, 25% educational + transportation, 20% healthcare. Range: 20%-31%, mean ~27%.

  • The workflow-not-task imperative:

    • “Nearly 90 percent of companies say they have invested in [AI], but fewer than 40 percent report measurable gains.” The gap is workflow integration, not technology adoption. Workflow-redesign vs task-automation framing — “the difference between offering employees access to a chatbot for ad hoc use and deploying custom agents alongside people in a reimagined process to approve, process, and manage loans more efficiently and deliver better customer service”.
    • 190 business processes mapped across 16 business functions; ~100 cross-cutting workflows + ~90 sector-specific workflows.

  • Four worked operational cases (Ch.3):

    1. B2B sales redesign (global tech firm, Exhibit 16) — 5 agents (prioritisation / outreach / customer response / scheduling / handoff) → 7-12% revenue increase + 30-50% time saved across sales roles. Future: coaching agent + admin agent.
    2. Utility customer operations (large utility, 7M+ support calls/year, Exhibit 17) — 4 agents (inbound call / intent identification / call scheduling / self-service) handle ~40% of all calls, resolve >80% without human involvement; avg cost per call cut ~50%; CSAT +6pp; humans handle complex/emotional/relationship-based issues. Long-term ambition: 80-90% of customer inquiries.
    3. Biopharma medical writing (global pharma, Exhibit 18) — 6 agents (clinical study planning / data mapping / report-drafting / validation / reviewing / submission draft) for clinical-study-report drafting → touch time for first human-reviewed drafts -60%, errors -50%, go-to-market accelerated by weeks.
    4. Regional bank IT modernisation (Exhibit 19) — 3+ agents (modernisation planning / assessment / functionality / coding / QA / testing) for legacy application modernisation; developers each coordinate 15-20 agents; code accuracy up to 70%; plans to reduce required human hours by up to 50%.

  • Leadership and management skill shifts (Exhibit 20, 10 leadership skills mapped to SCI position):

    • High SCI exposure (most change): Prioritisation, Decision-making
    • Medium SCI exposure: Planning, Coordinating, Budgeting, Accountability, Innovation
    • Low SCI exposure (least change): Coaching, Influencing, Mentorship — “AI will likely make managers more like coaches and orchestrators.”

  • Ch.4 leadership questions for business (6) — Are you reimagining your business for future value? Are you leading AI as a core business transformation (not delegating to IT)? Are you building a culture of experimentation and learning? Are you building trust and ensuring safety? Are you equipping your managers to lead teams of people, agents, and robots? Are you preparing your workers for new skills and roles?

  • Ch.4 leadership questions for institutions (3) — How can education and training keep pace? What systems are needed to ensure that transferable skills lead to new opportunities? How can local economies and communities respond? Historical anchors named: New Deal, GI Bill, COVID-19 institutional response, digital revolution through online learning + telehealth.

What was actually ingested

Full report + full technical appendix. No deferred content.

  • Report (agents-robots-and-us-skill-partnerships-in-the-age-of-ai-vf-final1.pdf): 60 pages, all 4 chapters + introduction + At-a-glance + Glossary + Acknowledgements + Endnotes (3 chapter-1 endnotes + 1 chapter-2 endnote + 6 introduction endnotes + 6 chapter-3 endnotes + 4 chapter-4 endnotes). All 20 numbered exhibits ingested. Both content sidebars (“How technology is advancing”, “Robots in the workplace”, “Framing the jobs debate as AI reshapes work”, “How we assess skill exposure to automation”, “How we estimate the economic value of AI”, “An early view of workflows across the US economy”). The 190+ workflow mapping across knowledge / frontline / production / cross-cutting domains is fully ingested.
  • Technical appendix (technical-appendix_agents-robots-and-us-vf-final.pdf): 6 pages, full ingest. Covers MGI’s automation and adoption models (BLS + O*NET + 18 human capabilities; sigmoidal S-curve diffusion; midpoint vs early/late scenarios), How we group occupations by technical automation potential (~800 occupations × ~2,000 DWAs; 55%-centric threshold), and How we quantify a skill’s potential to change (4 inputs: employment + DWAs + skills + automation model; OpenAI GPT-4o for ~3.4M skill→DWA labelling). The appendix’s three-example exhibit (Conflict resolution / Detail orientation / Ability to operate machines as people-led / shared / AI-led prototypes) operationalises the SCI classification rule visually.

Methodology lineage: MGI’s automation model was first developed in 2017 (Manyika et al., A Future That Works) and refreshed for this 2025 research on the US labour market with: (i) updated capability performance from new AI-expert survey, (ii) updated employment + wage data at BLS occupation level, (iii) new technical-frontier assessment for currently-demonstrated 2025 technologies, (iv) re-estimated solution timelines + costs + adoption curves with the midpoint scenario as the expected pace. Earlier MGI projection (June 2023, The economic potential of generative AI) put global potential at ~$26T without time-frame; this report puts it at $28.7T by 2030 — a tighter, time-bound projection.

Pre-flight checks (per CLAUDE.md §Verifying sources before ingest):

  • Scope — PDF reports 60 pages + 6-page appendix (pdfinfo confirmed); TOC references resolve within page count; no excerpt / sample / preview indication. Full ingest.
  • Identity — Cover names 7 authors + MGI + November 2025; PDF metadata title matches; PDF CreationDates (report Dec 8 2025, appendix Nov 25 2025) consistent with November 2025 cover branding. No filename-vs-content mismatch (agents-robots-and-us-...vf-final1.pdf is a normal McKinsey naming pattern; vf = version final; 1 likely indicates the first numbered revision after editorial pass). Earlier ingest precedent on MGI Race Takes Off used PDF CreationDate as date_published; this ingest follows the same convention but uses the appendix’s CreationDate (Nov 25) rather than the report’s (Dec 8), because the appendix CreationDate aligns more closely with the cover’s branded November 2025 release.
  • Honest scoping — length field states the actual scope: ~60-page report + ~6-page appendix, full ingest.

The Skill Change Index (SCI) as a substantive new framework

The SCI is the report’s central methodological contribution. The wiki has accumulated multiple sources on which skills matter in the AI era — but at impressionistic / vantage-specific altitudes:

SourceSkill-framework altitude
O DORAengineering-team core-skills (10 capabilities)
HBR IdeaCastneuroscience-vantage durable skills (self-regulation under uncertainty)
Lenny’spersonal-agency-as-durable-skill (curiosity, ownership, taste)
MGI / Yee et al. (this source)structural-modeling Skill Change Index across ~6,800 skills, ~1,800 occupations

The SCI is the wiki’s first labour-market-data-grounded systematic measure of which skills will change most and least under automation. Three properties make it useful as a durable-skills anchor:

  1. Coverage: 6,800 employer-cited skills from Lightcast across ~1,800 occupations. Not a selected list of “AI-era skills” — every skill in the US labour market with >5% job-posting prevalence.
  2. Time-weighted decomposition: each skill scored on the fraction of associated time spent on automatable work activities, weighted by occupation employment. Skills with high time-spent in non-automatable work activities are people-led (55% threshold) — even if some of their hours overlap with AI-capable tasks. The classification is therefore time-spent-weighted, not capability-binary.
  3. Scenario-bracketed: midpoint adoption scenario for 2030 (the report’s primary lens) vs early-adoption scenario (faster diffusion); SCI quartile values differ markedly across scenarios (midpoint top-quartile 33% vs early-scenario top-quartile 59%) — the report transparently shows the uncertainty band.

The SCI’s three buckets give the wiki a quantitatively-anchored complement to its prior durable-skills vocabulary:

  • Top quartile (highly exposed → will decline): SQL programming, inventory management, problem-solving (specific applications), invoicing.
  • Middle quartiles (will evolve): AI fluency itself, communication, customer relations, detail orientation, management, writing, quality assurance.
  • Bottom quartile (will endure): leadership, coaching, negotiation, management (general), good driving record.

How the report frames the jobs debate

The “Framing the jobs debate” sidebar (pp.12-13) is the report’s epistemological self-positioning. Four guiding questions:

  1. How close are AI agents and robots to matching all economically relevant human capabilities? — Not yet; AI “still lacks many distinctly human abilities, leaving ample room for human labor to thrive.” Multimodal models + advanced reasoning + fine motor skill + advanced social/emotional capabilities remain bottlenecks.
  2. Will a more AI-centric economy create enough jobs? — “This is beyond the scope of our analysis.” Names the open question rather than answering it. Cites BLS + WEF projections of continued employment growth + historical pattern (Industrial Revolution to internet); also flags the Acemoglu-Restrepo industrial-robotics work showing localised job losses and wage pressure.
  3. How might the composition of work change? — Names the flat college wage premium since ~2010 + plateaued knowledge-job salaries since mid-2024 as labour-market signals; new markets emerging in data-center construction + AI infrastructure maintenance + healthcare + personal services.
  4. Will we adapt fast enough? — Cites OECD on adult-learning + reskilling readiness; 77% of companies say they intend to launch upskilling/reskilling but follow-through is limited because employers recognise “workers often move on after gaining new skills”. Educational systems + social safety nets are the institutional dependency.

The report’s framing is explicitly avoidance of the AI-will-eliminate-jobs frame — it names that frame as the public-debate default and positions its own contribution as measuring skill change, not predicting job losses. This is the wiki’s most rigorously hedged structural source on the AI-labour question.

Workflow-redesign vs task-automation as the load-bearing prescription

Chapter 3’s central claim is operationally identical to:

  • Bain’s micro-productivity-trap (task gains don’t aggregate to firm value);
  • McKinsey’s “half-or-more of the secret sauce is organisational change, not technology implementation”;
  • AWS’s “AI bolted on is going to fail… ability to focus at a workflow level being the difference between success and failure”.

MGI’s contribution is the structural-modeling substrate for this prescription:

  • 190 workflows mapped across 16 business functions (commercial / functional / operational / sector-specific). The complete mapping is the wiki’s most granular publicly-available workflow taxonomy at MGI altitude.
  • The <40%-of-90% statistic — “Nearly 90 percent of companies say they have invested in [AI], but fewer than 40 percent report measurable gains” — is the workflow-redesign-disciplines-against-task-automation empirical anchor. The gap is workflow integration, not technology adoption.
  • The sector-specific concentration (60% of value in sector-specific workflows, 40% in cross-cutting) is the load-bearing prioritisation guidance for leaders: invest in sector-specific workflow redesign first, cross-cutting second.

The four worked cases (B2B sales / utility customer ops / biopharma medical writing / regional bank IT modernisation) are the operational analogs to Bain’s four-step transformation framework. The wiki now has multi-source cross-consultancy convergence on workflow-as-unit-of-AI-value-capture: MGI + McKinsey + Bain + AWS-advisory + Anthropic-engineering all naming the same prescription from independent vantages.

Leadership skills evolve too — managers as coaches and orchestrators

Exhibit 20 is the report’s contribution to leadership / management skill evolution. Mapped against the SCI:

  • High change: Prioritisation, Decision-making — agents will sequence tasks dynamically + simulate scenarios; humans will balance stakeholder needs + apply judgment to AI-simulated scenarios.
  • Medium change: Planning, Coordinating, Budgeting, Accountability, Innovation — agents reallocate resources + orchestrate workflows + monitor spending + generate audit trails + simulate prototypes; humans align stakeholders + resolve flagged conflicts + adjust priorities + interpret audit trails for integrity + test AI-generated concepts.
  • Low change: Coaching, Influencing, Mentorship — agents surface insights + analyse sentiment + analyse conversations for growth signals; humans tailor feedback + shape narratives + build relationships.

The pattern: managers shift from supervising people to orchestrating systems in which people, agents, and robots collaborate. MGI’s framing converges with Jassy 2025’s “changing role of managers” thesis at Amazon altitude and with Allen 2026’s builders-to-orchestrators role-shift at AWS-vendor altitude — three independent sources converging on the manager-as-orchestrator shape.

Why this matters to the wiki

This is the wiki’s most structural panoramic anchor on the AI-labour-and-skills question — the source most leaders would want on their desk when planning AI-era workforce strategy. Five anchoring contributions to the wiki’s existing knowledge graph:

  1. Numerical scaffold: 57% technical-automation potential; 65/35 nonphysical/physical work-hour split; 44/13 agent/robot capacity split; 75% AI-skill demand from 3 occupation groups; $2.9T US value at midpoint by 2030; 90% companies invest vs <40% report gains; 6.8× AI-fluency demand growth in 2 years; 34/21/30/<1/8/5/2 archetype shares.
  2. The Skill Change Index as a substantive new framework — first wiki source measuring skill-change-potential at labour-market-data-grounded altitude.
  3. The seven-archetype occupation taxonomy — the wiki’s first comprehensive prospective workforce-distribution frame; complements the micro-productivity-trap’s firm-level diagnosis with workforce-composition modeling.
  4. The workflow-redesign-vs-task-automation prescription at MGI structural altitude — joins the cross-consultancy convergence (Bain + McKinsey + AWS-advisory) on workflow as the unit of AI value capture.
  5. The MGI two-layer panorama — complementing MGI Race Takes Off (industry layer) with the workforce-skills layer. The same intellectual project at MGI editorial altitude, designed to be read in tandem.

The W&W tagging spans 12 cells — the broadest cell-coverage of any single source in the wiki — reflecting the report’s panoramic structural altitude. The report exercises virtually the entire W&W process model: digital-sensing through external-trigger framing + scenario-planning; digital-seizing through agent-economic modelling + workflow rapid-prototyping cases + strategic-agility under uncertain timelines; digital-transforming through internal-structure redesign (the 4 cases) + digital-maturity prescriptions (AI fluency 6.8× demand growth); strategic-renewal through business-model value-capture modelling + organisational-culture (the Ch.4 culture of experimentation question); and the full contextual ring through trigger / enabler / barrier identification.

Linked entities and concepts

Existing wiki entities mentioned:

  • McKinsey & Company — the parent firm; this report’s MGI is the firm’s research arm. Endnotes cite multiple recent McKinsey companion publications: The state of AI in 2025: Agents, innovation, and transformation (Nov 5 2025, cited Ch.4 endnotes 23-24), Performance through people (Feb 2 2023, endnote 25), The agentic organization (Sept 26 2025, endnote 26), The economic potential of generative AI (June 2023, endnote 3 of methodology sidebar), In search of cloud value: Can generative AI transform cloud ROI? (Nov 2023, endnote 7), McKinsey technology trends outlook 2025 (July 2025, endnote 2 of How-Tech-Is-Advancing sidebar), A leap in automation: The new technology behind general-purpose robots (July 2025), Will embodied AI create robotic coworkers? (June 2025), Humanoid robots: Crossing the chasm from concept to commercial reality (October 2025). Open follow-ups; not separately ingested.
  • Anthropic — endnote 1 of Ch.2 (“Claude 3.5 Sonnet,” Anthropic, June 2024) as the canonical multistep-task reasoning anchor.

New entity promoted in this ingest (3rd-mention rule):

Concepts substantively touched in this ingest:

  • automation-vs-augmentation — 57%/43% potential, 44%/13% agent/robot split, 7-archetype taxonomy as the wiki’s most granular automation-vs-augmentation structural anchor.
  • ai-employment-effects — shift-not-elimination framing; 72% shared-skills finding; SCI as quantitative complement.
  • durable-skills — SCI as systematic measure; 8 high-prevalence transferable skills; bottom-quartile-endures + middle-quartile-evolves + top-quartile-declines.
  • micro-productivity-trap — the <40%-of-90% statistic + workflow-redesign-not-task-automation prescription at MGI structural altitude.
  • enterprise-ai-adoption — workflow-as-unit-of-redesign; sector-specific (60%) + cross-cutting (40%) value concentration; $2.9T US / $28.7T global scaffold.
  • dynamic-capabilities — workflow redesign as digital-transforming/redesigning-internal-structures empirical anchor; the report’s 12-W&W-cell coverage is the broadest in the wiki.

Dangling first-mentions (single-source author appearances, deferred per §Author-entity promotion):

  • Lareina Yee — MGI director, senior partner, Bay Area office.
  • Anu Madgavkar — MGI partner, New Jersey office (also named in the MGI Race-Takes-Off org-chart context).
  • Sven Smit — MGI chairman, senior partner, Amsterdam office (also named in MGI Race-Takes-Off org-chart context, where listed as the current MGI chair).
  • Alexis Krivkovich — McKinsey senior partner, Bay Area.
  • Michael Chui — QuantumBlack senior fellow, Bay Area (well-known McKinsey AI thought leader).
  • Maria Jesus Ramirez — MGI senior fellow, Bay Area.
  • Diego Castresana — McKinsey engagement manager, New York.
  • Sir Christopher Pissarides — Nobel laureate (Economics 2010), Regius Professor of Economics at LSE; MGI academic adviser.
  • Matthew J. Slaughter — Paul Danos Dean of the Tuck School of Business at Dartmouth; MGI academic adviser.
  • Luca Vendraminelli — postdoctoral researcher at Stanford Digital Economy Lab + Stanford HAI; named as research contributor to this report.

Cited authors / sources not separately ingested (open follow-ups): Marc Benioff (Time 2024, Salesforce framing of unlimited age), Ivan Solovyev + Shrestha Basu Mallick (Google, Gemini 2.0 announcement), Eric Christensen et al. (Projected US Radiologist Supply 2025-2055, NIH Feb 2025), Heinz-Peter Schlemmer (Will radiology sink or soar, National Library of Medicine July 2025), Steve Lohr (NYT, May 14 2025), Jeffrey Lin (Review of Economics and Statistics, 2011), Danny Driess et al. (ICRA 2021), Paul Gaggl et al. (NBER 26477, Nov 2019), Daron Acemoglu + Pascual Restrepo (NBER 23285, March 2017; World Bank July 2019), Stephen J. Klein + Nathan Rosenberg (World Scientific 2009), Chad P. Brown + Caroline Freund (Peterson Institute Jan 2019). The Brynjolfsson Canaries in the Coal Mine is already ingested as 2026-04-28-brynjolfsson-canaries-coal-mine — endnote 5 of the Framing-the-Jobs-Debate sidebar cites it directly.

Notes on source quality

kind: report. Full ingest of both the report and the technical appendix — the wiki’s most comprehensive structural-AI-labour anchor. The report is a McKinsey Global Institute flagship publication; MGI publications are “independent and have not been commissioned or sponsored in any way by any business, government, or other institution” per the report’s acknowledgements. Methodology is transparently disclosed (BLS + O*NET + Lightcast + GPT-4o + manual 1,000-cell validation template; sigmoidal-curve diffusion modelling; midpoint vs early/late scenario brackets). Authorial altitude is MGI partner + senior fellow (the highest McKinsey research altitude), with academic-adviser sign-off from Pissarides (LSE Nobel-laureate economist) + Slaughter (Tuck Dean). The combination of methodological transparency + tier-1 academic sign-off + the MGI publication-process discipline puts this at the highest source-reliability tier in the wiki — comparable to AI Index 2026 (Stanford HAI), Warner & Wäger 2019 (peer-reviewed International Journal of Information Management), and Krakowski et al. 2025 (peer-reviewed Management Science).

Direct verbatim quotation from the PDF is reliable (formal report typography, edited prose). The 20 numbered exhibits are visually complex (multi-panel charts, sector-by-domain heatmaps, archetype square-grid visualisations) and lossy when reduced to markdown tables — body prose preserves the headline numbers; readers needing the visual structure should consult the source PDF directly. The technical appendix’s 3-skill exhibit (people-led / shared / AI-led prototypes) is also visually structured and is described in prose in the body.

Relationships

Supports

  • Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence — **Descriptive labour-market data + prospective structural model on the AI labour question.** Brynjolfsson et al. (*Canaries in the Coal Mine*, Aug 2025): six empirical facts from US payroll data showing the augmentation/automation cut already producing labour-market signals — *augmentative AI use does not displace labour while automative use does*. MGI (Yee et al., Nov 2025) supplies the prospective frame: technical automation potential covers **~57% of US work hours** at midpoint scenario; the labour effect is shifted-not-deleted (7 archetypes from people-centric to robot-centric); 72% of today's skills are required for both automatable and non-automatable work, so most skills *evolve* rather than disappear; ~$2.9T US economic value by 2030 conditional on workflow-not-task redesign. The MGI/Brynjolfsson pair is the wiki's strongest **descriptive-empirical + structural-prospective** anchor on the AI labour question — Brynjolfsson establishes that the augmentation/automation cut is *operationally meaningful in current data*; MGI maps *where the cut is heading by 2030* with occupation-level granularity. Both reject the *AI-will-eliminate-jobs* framing in favour of *AI-will-reshape-which-skills-are-applied-where*.
  • Human-Centered Artificial Intelligence: A Field Experiment — **Individual-level tailoring + firm-level workflow redesign as paired levers.** Krakowski et al. (*Management Science* 2025, Nordic pharma DiD, N=72): untailored AI augmentation gradually decreases market share vs the legacy-IT control; tailored AI per Kirton 1976 cognitive style produces positive effects across four interaction-design parameters (*work procedure / decision-making authority / training / incentives*). MGI (Yee et al., Nov 2025): the workflow-redesign-not-task-automation prescription — *<40% of companies that have invested in AI report measurable gains* despite ~90% investing; the gap is workflow integration, not technology adoption — *'the difference between offering employees access to a chatbot for ad hoc use and deploying custom agents alongside people in a reimagined process'*. Krakowski's cognitive-style-tailoring is the **individual-interaction-design** layer; MGI's workflow redesign is the **firm-process-design** layer. Same load-bearing claim — *tailored intervention beats blanket prescription* — at two stack altitudes. Both anchor the W&W `digital-transforming/redesigning-internal-structures` cell with empirical evidence (Krakowski) and prospective structural modeling (MGI).
  • The race takes off in the next big arenas of competition (MGI 2026 update) — **Same-publisher (MGI) complementary-panoramas pair.** Russell + Bradley et al. (MGI, March 2026, *Race Takes Off in the Next Big Arenas of Competition*): industry-and-competition layer — 18 arenas of value migration, 80% of consequential AI-era value flows through 10–20 arenas, AI as cross-arena accelerant. Yee et al. (MGI, Nov 2025, *Agents, Robots, and Us*): workforce-and-skills layer — 7 occupation archetypes, Skill Change Index, $2.9T economic value contingent on workflow redesign, the agents-and-robots collaboration framing. The two MGI flagships form a **two-layer panorama** of the AI-and-economy question: where will the value migrate (Race Takes Off) × who and what will do the work that captures it (Agents, Robots, and Us). Same publisher, same intellectual project, **6-month delta**, designed-to-be-complementary at the MGI editorial level. **Recurring MGI partner overlaps**: Anu Madgavkar appears as MGI-partner-overlap (Race Takes Off acknowledges her in MGI org chart; Agents-Robots-and-Us lists her as second author). MGI directors **Chris Bradley** (Race Takes Off co-author) and **Sven Smit** (Agents-Robots author) are the cross-report anchors. The MGI Race Takes Off ingest's open follow-up — *update [[McKinsey & Company]] with MGI sub-section* — is fulfilled by this ingest's [[McKinsey Global Institute]] entity-page promotion.
  • How McKinsey Plans to Survive AI (and Reinvent Consulting) — **McKinsey-firm + MGI consultant-altitude pair on the workflow-redesign prescription.** Sternfels (McKinsey Global Managing Partner, HBR IdeaCast Feb 2026): *'half, if not more, of the secret sauce is organizational change as opposed to technology implementation'*; flatter orgs + workflow consolidation; consulting-engagement-evidence vantage. Yee et al. (MGI, Nov 2025): the same prescription with structural-modeling evidence — *'realising these gains requires more than automating individual tasks. It will mean redesigning entire workflows so that people, agents, and robots can work together effectively'*; the **<40% of companies report measurable gains** statistic is the workflow-redesign-disciplines-against-task-automation empirical anchor; 190 business processes mapped across 16 functions; ~60% of potential gains concentrated in sector-specific workflow domains. McKinsey-firm leadership (Sternfels) supplies the operating-doctrine narrative; MGI (Yee et al.) supplies the structural-modeling substrate. Two McKinsey altitudes — practitioner-firm + research-institute — converging on the same workflow-redesign-not-task-automation thesis.
  • How to Move from AI Experimentation to AI Transformation — **Cross-consultancy convergence on workflow-as-unit-of-AI-value-capture.** Dutt + Chatterji (Bain/OpenAI Apr-May 2026): the **micro-productivity-trap** thesis — task-level AI gains failing to translate to firm-level results; two named lock-ins (*offering*, *process*); escape via the four-step transformation framework; 10–25% Bain client EBITDA gains contingent on transformation. Yee et al. (MGI, Nov 2025): the same task-vs-workflow distinction at structural-modeling altitude — *'applying AI to individual tasks within these legacy processes is unlikely to deliver the productivity gains now possible'*; ~$2.9T US economic value conditional on workflow redesign; 190 workflows mapped as the unit-of-analysis (16 functions across knowledge / frontline / production services). Bain + McKinsey/MGI **independent consultancies converging** on the workflow-as-unit-of-AI-redesign discipline — the wiki's strongest cross-consultancy structural evidence that the task-vs-workflow distinction is operationally load-bearing, not consulting-firm-specific framing. MGI's 4 worked cases (B2B sales, utility customer ops, biopharma medical writing, regional bank IT modernization) are the operational analogs of Bain's four-step transformation framework.
  • Artificial Intelligence Index Report 2026 — **Two 2026-era panoramic AI reports at complementary layers.** Stanford HAI / Maslej et al. (*AI Index Report 2026*, Apr 2026): capability-and-investment layer — frontier-model benchmarks, training compute trends, investment patterns, geographic distribution. MGI / Yee et al. (*Agents, Robots, and Us*, Nov 2025): workforce-and-skills layer — technical automation potential by occupation, Skill Change Index, $2.9T economic value, workflow redesign as value-capture mechanism. Two **independent annual-panorama reports** at adjacent stack layers — AI Index measures *what AI can do*, MGI measures *what people will do when AI can do more*. The reports are designed to be read in tandem by leaders and policymakers; both 2026 publications are systematic structural anchors at their respective altitudes.
  • Generative AI at Work — **Field-experiment workflow productivity evidence + structural-modeling workflow prescription.** Brynjolfsson + Li + Raymond (*Generative AI at Work*, INFORMS 2026, customer-support RCT, N=5,179 agents at a Fortune 500 firm): GenAI assistant access drove **14% mean productivity gain** (calls resolved per hour), concentrated in **novice and low-skilled workers** (+34%) with no measurable effect on experienced workers; improved customer-sentiment ratings; reduced resignations. MGI (Yee et al., Nov 2025): the customer-operations case study (utility company, 7M+ support calls/year) operationalises the same human–agent collaboration shape — inbound call agent + intent identification + call scheduling + self-service agent handling ~40% of calls, resolving >80% without human involvement; **avg cost per call cut ~50%**, CSAT +6pp, human reps now handle complex/emotionally-sensitive issues. Brynjolfsson supplies the empirical-RCT-evidence layer; MGI supplies the structural-prescription + worked-case-illustration layer. Same workflow shape (AI handles routine, humans focus on complex), two altitudes.

Graph View

Backlinks