Nobody Knows How to Manage Robots at Work

McKinsey’s landmark AI workforce study projects $670 billion in robot automation value, but can’t explain how production workers will actually coordinate with autonomous machines.

free link

McKinsey’s November 2025 report “Agents, robots, and us: Skill partnerships in the age of AI” projects $2.9 trillion in annual US economic value from AI-powered automation by 2030. Of that total, $670 billion, nearly a quarter, comes from robots automating physical work in manufacturing, construction, and agriculture.

The report maps seven distinct workforce archetypes, analyzes 6,800 skills across 11 million job postings, and provides detailed implementation frameworks for how organizations should redesign workflows around human-AI collaboration.

The Implementation Asymmetry

McKinsey’s agent transformation framework is remarkably detailed. Knowledge workers, the report explains, will shift from task execution to orchestration — directing teams of 20+ AI agents, validating outputs, managing escalations, and handling exceptions.

• A medical writer moves from drafting clinical documents to reviewing AI-generated content and ensuring regulatory compliance.
• A software developer shifts from writing code to architecting agent workflows and quality assurance.
• Sales representatives transition from prospect research to relationship management while agents handle lead qualification and proposal generation.

Each archetype comes with time reallocation patterns. Medical writing: 30–70% time savings. Software development: agents handle 40–60% of coding tasks. Customer service: humans focus on complex escalations while agents manage routine inquiries.

The framework provides workforce planning guidance, skill development pathways, and change management approaches.

Now examine the robot projections. Manufacturing workflows are 32% automatable by robots, construction 38%, agriculture 43%. These percentages apply to 26 million US workers in occupations McKinsey classifies as robot-centric (production operators, machine setters, agricultural workers) or people-robot hybrid (maintenance technicians, construction specialists).

The potential economic value is quantified with the same precision as agent automation: manufacturing contributes $210 billion, construction $156 billion, agriculture $89 billion to the $670 billion total.

But there’s no implementation layer beneath these numbers.

No case studies of assembly line operators shifting to robot orchestration. No workflow redesign patterns for construction sites integrating autonomous equipment. No time reallocation analysis for agricultural workers managing robotic harvesting systems.

The report offers a single illustrative example of a solar facility operation combining drones, autonomous rovers, and field technicians; but even this reinforces the pattern.

The human technician orchestrates two AI agents managing system performance and energy optimization. The robots operate autonomously under agent supervision. The framework remains agent-centric even when robots are involved.

The asymmetry suggests a different question than “how do we implement robot automation?”

It suggests: why isn’t robot implementation ready for consulting engagements yet?

The Three Structural Gaps

Gap One: The Economics Don’t Work

McKinsey’s robotics sidebar provides the answer explicitly. Current humanoid robots cost $150,000 to $500,000 per unit. For mass adoption in manufacturing, construction, and agriculture, per-unit costs need to fall to $20,000 to $50,000 — a 75–90% reduction.

Other constraints compound the problem: battery life limits operation to 2–4 hours per charge, safety systems for human-robot collaboration remain unsolved, and fine motor control for complex manipulation tasks doesn’t yet meet reliability thresholds.

This explains the absence of implementation patterns. Organizations aren’t deploying production robots at enterprise scale because the business case doesn’t close. When an assembly line worker earns $42,000 annually (the average wage for robot-centric roles in McKinsey’s archetype analysis) but the robot replacing them costs $300,000 with 3-hour operational windows and requires extensive safety infrastructure, the ROI timeline extends beyond strategic planning horizons.

Agent economics work differently. A software agent incurs marginal compute costs and scales near-instantly. API calls cost fractions of pennies. Inference runs continuously without hardware replacement cycles. The asymmetry between agent and robot deployment economics creates an asymmetry in implementation maturity — knowledge work transformation is happening now because the infrastructure already exists, while production work transformation remains theoretical because the cost structure hasn’t reached viability.

Gap Two: The Pay Hierarchy Reveals Priority Signals

McKinsey’s seven workforce archetypes come with average wage data that tells a story about client priorities:

• Agent-centric roles: $70,000 average annual pay (accountants, software developers, lawyers)
• People-agent roles: $74,000 (sales representatives, teachers, engineers)
• Robot-centric roles: $42,000 (stockers, welders, machine operators)
• Agent-robot roles: $49,000 (machine setters, production workers)
• People-robot roles: $54,000 (maintenance technicians, construction workers)

The highest-paid workers facing automation are knowledge workers whose transformation receives detailed orchestration frameworks. The lowest-paid workers facing automation are production workers whose transformation receives economic projections without implementation guidance. This isn’t coincidental — it reflects where consulting engagements focus. Enterprise clients purchasing $40 million McKinsey studies are primarily concerned with knowledge worker productivity because that’s where competitive advantage concentrates in services, technology, and financial services sectors.

Manufacturing, construction, and agriculture employ different buying patterns. Production optimization projects historically focus on equipment efficiency and supply chain velocity, not workforce transformation strategy. When robots eventually reach deployment economics, the implementation frameworks will likely come from manufacturing engineering firms and equipment vendors, not management consultancies. But that creates a dangerous gap: knowledge workers get transformation pathways that emphasize orchestration and skill evolution, while production workers may face automation that emphasizes labor substitution without parallel workforce development infrastructure.

The 5% of workers in people-agent-robot roles 9the most complex orchestration challenge, involving human judgment, software intelligence, and physical automation simultaneously0 receive essentially no guidance. These 7.5 million workers in transportation, agriculture, and food service sit at the convergence of all three automation modes, yet McKinsey’s framework treats them as a residual category rather than a design priority.

Gap Three: The Vocabulary Doesn’t Exist

Chapter 2 of McKinsey’s report analyzes the eight high-prevalence skills that remain essential across all industries: communication, management, operations, problem solving, leadership, detail orientation, customer relations, and writing. These skills are transferable, appearing in both automatable and non-automatable work, which McKinsey argues makes them the foundation for workforce adaptability as AI adoption accelerates.

But examine what these skills describe: cognitive and social capabilities. Communication, writing, problem solving, leadership — these are the competencies of knowledge work. They map cleanly onto agent orchestration because agents perform cognitive tasks that complement these human skills. A manager directing an AI agent uses communication, problem solving, and leadership in ways that parallel managing human teams. The vocabulary translates.

McKinsey identifies a separate skill category called “working with machinery and specialized equipment” that encompasses the physical competencies required in manufacturing, construction, and agriculture. This category gets mentioned in the skill distribution analysis but receives no transformation framework equivalent to the agent orchestration model.

The report describes how writing shifts from “drafting content” to “refining AI outputs,” how problem solving shifts from “analyzing data” to “interpreting AI findings,” how management shifts from “tracking metrics” to “coaching hybrid teams.” These are concrete operational changes that CTOs can translate into workforce development programs.

There’s no parallel framework for how “operating industrial machinery” shifts when robots handle physical manipulation, or how “construction assembly” evolves when autonomous equipment performs tasks under human direction, or how “agricultural equipment operation” transforms when robotic systems manage harvesting and workers orchestrate fleet behavior. The language for robot orchestration competencies hasn’t been developed — or more precisely, hasn’t been abstracted from manufacturing engineering into the strategic workforce planning vocabulary that consulting firms deploy.

This linguistic gap compounds the implementation void. Even if robot economics improved tomorrow, organizations lack the conceptual frameworks to design human-robot skill partnerships the way they’re now designing human-agent partnerships. Knowledge work transformation has a language — orchestration, validation, exception handling, escalation management. Production work transformation has equipment specifications and efficiency metrics, but not yet a parallel vocabulary for how human expertise evolves in collaborative robot environments.

The Missing Architecture: Production Work Orchestration

What would robot orchestration frameworks actually look like if they existed? The solar operations example provides one data point: field technicians supervise autonomous drones and rovers while coordinating with AI agents managing system performance.

But this is maintenance work isolated to episodic intervention on equipment that operates independently. It’s not assembly line production where human-robot collaboration happens continuously, or construction where work cell configuration changes by project phase, or agriculture where environmental variability requires real-time adaptation.

The architectural pattern for knowledge work orchestration is emerging clearly: humans operate at decision surfaces where they frame problems, set parameters, validate outputs, and handle exceptions while agents execute structured workflows.

The human role combines:

• domain expertise (understanding what good output looks like),
• process design (architecting effective agent workflows),
• quality control (catching errors that automated systems miss).

This preserves human judgment and contextual understanding while leveraging machine scale and consistency.

Manufacturing workflows are spatially constrained — work happens at physical stations with specific material flows and safety requirements. Human-robot collaboration must account for shared workspace dynamics, real-time coordination of physical actions, and exception handling that involves both digital and physical intervention. A production worker can’t simply “validate” robot output the way a medical writer validates text — they may need to physically adjust materials, recalibrate equipment, or directly intervene in physical processes.

Construction sites present even more complexity. Work cells reconfigure constantly as projects progress. Equipment moves between locations. Environmental conditions vary. The orchestration model can’t be static workflow optimization — it requires dynamic replanning as constraints shift. Agricultural operations add biological variability: crops don’t standardize the way manufactured parts do, weather creates unpredictable operational windows, and autonomous equipment must adapt to terrain variations that knowledge work never encounters.

These aren’t implementation details. They’re fundamental architectural differences between cognitive and physical work that require distinct orchestration models.

• Knowledge work happens in digital spaces where state is explicit, changes are reversible, and coordination happens through data interfaces.
• Physical work happens in three-dimensional space where state is distributed across materials and equipment, many changes are irreversible, and coordination requires spatial awareness and timing.

Enterprise architects planning automation strategies need both models. The agent orchestration framework applies to knowledge work. The robot orchestration framework, which doesn’t yet exist in strategic planning vocabulary, needs to account for spatial constraints, physical safety, real-time coordination, and the different competencies required when humans and robots share workspace rather than just share information flow.

Strategic Implications for Enterprise Leaders

The $670 billion robot value projection isn’t wrong, it’s premature. The automation potential exists, the economic value is calculable, but the implementation infrastructure isn’t ready. For CTOs and enterprise architects, this creates three strategic priorities that differ from knowledge work transformation:

First, separate robot business cases from agent business cases. Agent automation has reached implementation maturity with proven ROI patterns and established orchestration frameworks. Robot automation remains in the pilot phase with cost structures that don’t yet support enterprise scale.

Organizations should invest aggressively in agent workflows while treating robot deployment as exploratory R&D rather than operational transformation. The timeline divergence matters: knowledge work orchestration becomes a competitive capability in 2025–2027, while production work orchestration likely reaches enterprise readiness in 2028–2030 or later, depending on hardware cost curves.

Second, don’t wait for consulting firms to develop production orchestration frameworks. The vocabulary gap and client priority patterns suggest management consultancies won’t lead here. Manufacturing engineering firms, equipment vendors, and production operations teams will develop the first real-world patterns through direct experimentation.

Enterprise leaders should build internal capability by partnering with equipment suppliers on pilot programs, documenting orchestration patterns as they emerge, and developing workforce competency models specific to their production environments. The knowledge work playbook won’t translate directly; manufacturing, construction, and agriculture need domain-specific orchestration architectures.

Third, address the class implications explicitly. The 66% wage premium for agent-centric over robot-centric roles reflects current market structure, but it creates a dangerous precedent if knowledge workers receive orchestration-based transformation while production workers face substitution-based automation. Organizations have agency in how they structure these transitions.

Workforce development programs should emphasize skill portability for production workers the same way McKinsey emphasizes it for knowledge workers; the 72% of skills that apply across automatable and non-automatable work exist in manufacturing and construction, not just software and finance.

The absence of robot implementation frameworks in McKinsey’s landmark study isn’t a critique of the research, it’s a signal about the current state of automation readiness.

Agent transformation is happening now because the infrastructure matured. Robot transformation is projected because the potential is clear, but implementation will require different economics, different orchestration models, and different workforce development approaches than knowledge work received.

The $670 billion opportunity exists.

The implementation framework doesn’t, yet.