Publication Notice: 'Human and AI/Robot Collaboration / AI Employees / Labor Portfolio in the AI Era / Talent Intelligence / Native AI Organization Whitepaper 2026 Edition'

The Institute of Next Generation Social System R&D (INGS) announced the publication and overview of the 'Human and AI/Robot Collaboration / AI Employees / Labor Portfolio in the AI Era / Talent Intelligence / Native AI Organization Whitepaper 2026 Edition' on April 1, 2026.

## Information from the Whitepaper Editorial Team

## Key Message

The era has arrived where AI employees and digital workers are managed not as 'tools' but as 'workforce'.

The WEF's 'Future of Jobs Report 2025' predicted that 40% of job skills across all industries would be transformed by 2027, simultaneously recording a surge in 22 AI-related professions and a displacement risk of 78 million jobs.

In an experiment conducted by Harvard Business School on 776 employees of P&G, teams collaborating with AI outperformed solo workers with statistical significance. As Professor Karim Lakhani states, 'AI is no longer a tool, but a teammate.' Organizations that fail to make this cognitive shift will lose their 'talent' to competitors.

As demonstrated by next-generation models such as Google DeepMind Genie 2/3, Physical Intelligence π0/π0.5, WorldLabs' Marble3D, and Meta V-JEPA 2, AI has begun to transcend digital spaces and embed itself in physical processes. The philosophy of 'human-centric automation' championed by Industry 5.0 is now taking concrete shape through on-site implementations of the international standard ISO/TS 15066 (safety requirements for collaborative industrial robot systems) and 4th-generation industrial robots.

However, more important than the numbers is the fact that the 'space' for collaboration has changed. Collaboration with AI is no longer a phenomenon confined to digital screens.

On the factory floor, Cobots (collaborative robots) stand next to humans under safety designs compliant with ISO/TS 15066, and in warehouses, AMRs/AGVs dynamically switch between autonomous decision-making and routing exceptions to humans via AITL (Agent-in-the-Loop) design.

In operating rooms, AI provides diagnostic support while guaranteeing the physician's final judgment through HITL (Human-in-the-Loop). The 'human-centric automation' advocated by Industry 5.0 has become a matter of on-site implementation rather than mere philosophy in 2026.

The design of HITL (Human-in-the-Loop) and AITL (Agent-in-the-Loop) is the only key enabling production deployment—as data showing that 'only 6% of companies fully trust agents' indicates, AI deployment without human involvement design will fail.

And now, consulting firms themselves are transforming into the largest implementers of AI.
McKinsey QuantumBlack is deploying AI agents horizontally on a scale of 25,000 people, and Accenture is incorporating 11,000 AI employees (digital workers) into its consultant workforce.

PwC provides a multi-agent foundation called Agent OS to its clients, and Deloitte, KPMG, and IBM are deploying HITL/AITL design as standard services through partnerships with Google Cloud and watsonx, respectively.

Major consulting firms are now shifting their roles from advisors to 'co-implementers'. This change signifies the end of the consulting firm as a mere outsourcing destination for AI strategy, marking the beginning of a new collaborative model as joint design partners.

Thus, this whitepaper poses the question that remains after the discourse of 'using AI/robots' has become obsolete. The question to ask is the design philosophy of HITL/AITL itself: where should humans make judgments, and where should tasks be delegated to agents or collaborative robots?

## Usage Scenarios

This whitepaper is written for organizations suffering from a lack of 'decision-making structure' rather than a lack of 'knowledge' regarding AI. The following scenarios represent the frontlines where these decisions carry the most practical weight.

### ① When transitioning AI agents to production deployment is an urgent task

As Gartner research shows, by 2026, 40% of enterprise applications have reached the stage of incorporating task-specific AI agents. However, at the same time, as recorded by Dynatrace, 50% of companies conducting PoCs for agent AI cannot measure ROI, and 74% do not even have a policy for ROI measurement.

This whitepaper unravels the four structural factors hindering graduation from PoC, centered around the 5-stage AI maturity model (Augmented Search -> Custom Tool Creation) shown by research from MIT CISR, Automation Anywhere, and WiserMethod. It systematically analyzes the conditions for EBIT improvement (+4.7% for Stage 3, +13.9% for Stage 4) achieved by transitioning from Stage 2 to Stages 3 and 4.

It covers the necessary technology layers for enterprise production deployment, from establishing MCP (Model Context Protocol) and implementing Agentic RAG, to design guidelines for multi-agent collaborative architectures.

### ② Manufacturing, logistics, and medical sites requiring physical human-robot collaboration design

Cobots (collaborative robots) are no longer 'manufacturing machines' but 'new colleagues powered by AI'. For DX leaders in manufacturing, implementing Cobots is no longer an issue of 'automation investment' but of 'redesigning the labor portfolio'.

4th-generation Cobots compliant with ISO/TS 15066 have reduced physical distance to humans to zero using force sensors, AI vision, and real-time safety protocols. AMRs and AGVs dynamically utilize autonomous judgments via AITL and exception handling via HITL, fundamentally transforming personnel efficiency in warehouses.

Real-time AI assistance for on-site operators leveraging AR (Augmented Reality) simultaneously improves the speed and accuracy of judgments in quality inspection, equipment maintenance, and process management.

Furthermore, this whitepaper details the business applicability of Physical AI and world model technologies pioneered by NVIDIA Cosmos, Google DeepMind Genie, Physical Intelligence π0 series, and Meta V-JEPA 2, along with the robot deployment cost revolution brought by Sim-to-Real technology, providing the technology map for the next 5 years that decision-makers in manufacturing need to know.

### ③ When considering the business application of Physical AI / World Model technology

The emergence of NVIDIA Cosmos, Google DeepMind Genie, Physical Intelligence π0 series, and Meta V-JEPA 2 is fundamentally rewriting the concept of 'AI for operating robots'.

Due to the rapid evolution of Sim-to-Real technology, the cost and time required to transfer simulation learning in digital spaces to physical robots have drastically decreased.

This whitepaper provides an analytical framework for evaluating the current state of these next-generation physical AI technologies and the feasibility of integrating them into existing production, logistics, medical, and infrastructure management facilities within the context of HITL/AITL design.

### ④ When HR strategy and organizational design for a hybrid workforce (Human + AI Agent + Collaborative Robot) are needed

The displacement of 78 million jobs and the creation of 19 million new roles warned by the WEF in 2025 are occurring simultaneously.

The problem is not 'whether jobs will disappear', but the speed of designing 'which roles to redefine, how, and what to teach to whom'.

Warehouse workers share 'role allocation protocols' with AMRs, and quality control technicians are redefined as HITL monitors for AI vision systems. Manufacturing maintenance engineers begin to function as 'verifiers' who validate AITL judgments in predictive maintenance by AI.

Workday's Agent System of Record provides a foundation to register, evaluate, and evolve AI agents as talent management subjects, while Salesforce Career Connect (by Adecco × Salesforce) and Adecco 'r.Potential' integrally design skill-based hiring and internal career mobility.

AI coaching platforms like CoachHub, MY Pi, TalentLMS, and Disprz achieve an organization-wide elevation of AI literacy at the scale of 1:1 digital coaching, as demonstrated by the deployment of GPTs to 18,000 BCG employees.

This whitepaper explains exactly which HITL design requirements Article 14 of the EU AI Act (human oversight obligation) imposes in the HR, manufacturing, and medical sectors, and provides guidelines for utilizing compliance not as a 'constraint' but as a 'baseline for trust design'.

### ⑤ When building an integrated operating model for 2026-2030 is set as a management agenda

The conclusion of this whitepaper is presented not as a 'future prediction' for 2030, but as a 'current structural decision' that will be too late if not designed now.

Building an autonomous AI architecture based on the MAPE-K (Monitor-Analyze-Plan-Execute-Knowledge) framework, Minimum Viable Organization (MVO) design principles, and AI compound interest structures (continuous accumulation of domain-specific protocols and proprietary intelligence)—these are not challenges for the technology department, but core themes of corporate strategy.

While overseeing the technological horizon including Web 4.0, AR/VR/MR integration, IoT/OT fusion, and quantum computing (IBM Qiskit AI), the roadmap for 2026-2030 is systematically shown through a 'two-axis compass' intersecting the Gartner 5-stage evolution model and the BCG 3-tier transformation roadmap.

### ⑥ Management Strategy / AI Investment Decision Scenarios

─ Companies can formulate their own AI investment priorities and company-wide rollout roadmaps based on the reality of an 88% AI usage rate, the structural factors of the 'PoC Trap', BCG's impact gap theory, and the Gartner 5-stage evolution model.

HR / Talent Strategy Planning Scenarios
─ It can be directly utilized for the 40% skill transformation indicated by the WEF 'Future of Jobs 2025', the design of skill-based hiring, internal career mobility, and performance review automation, as well as the selection of talent intelligence platforms.

DX / Digital Workplace Design Scenarios
─ It can be used as an implementation guide for MCP (Model Context Protocol), Agentic RAG, multi-agent collaborative architectures, and over 20 AI workflow patterns (HITL verification, parallelization, loop-type self-healing, etc.).

Compliance / Governance Response Scenarios
─ Readers can obtain systematic information on a gap analysis framework for EU AI Act compliance, NIST AI RMF, SOC 2 Type 2 compliance, and audit trail design for agent behavior.

### ⑦ When wanting to leverage AI deployment cases of consulting firms for internal strategy planning

Major global consulting firms have now transformed from 'those who talk about AI' to 'entities that implement AI on a large scale'.

McKinsey Agents-at-Scale and QuantumBlack have proven company-wide rollouts equivalent to AI maturity Stage 4 within their own operations and are offering them externally as reproducible implementation patterns.

BCG has accumulated quantitative evidence of productivity improvement through ChatGPT Enterprise (18,000-person scale) and the development of dedicated GPTs like Deckster and GENE.

PwC Agent OS is deployed to clients as a 4-layer multi-agent foundation, and KPMG × Google Agentspace has made cross-agent collaboration via the Agent2Agent protocol a standard implementation.

IBM watsonx + IBM Garage has systematized HITL/AITL design in healthcare, HR, and manufacturing domains as transition patterns from PoC to production, and NTT DATA and NRI have published AI CoE design guidelines to break through the unique barriers of the Japanese market regarding 'PoC to PoC'.

This whitepaper cross-analyzes the AI implementation architectures and deployment methods of these firms, presenting them in a structure that can be directly utilized as 'reference models' for formulating one's own AI strategy.

## Action Plan / Core Recommendations

The behavioral guidelines for management and organizations presented in this whitepaper are summarized in the following five points.

### ① Identify the '4 walls' of graduating from PoC and strive for a breakthrough

The EBIT +4.7% resulting from the transition from Stage 2 to Stage 3 shown by MIT CISR data is an issue of organizational commitment rather than technical maturity. Among the four structural factors constituting the PoC trap (absence of ROI measurement, lack of HITL design, absence of cross-organizational governance, and disconnected KPIs), identify the factors applicable to your company using the diagnostic framework in this whitepaper, and formulate a transition plan linked to the authority design of an AI CoE (Center of Excellence) within this term.

### ② Fully map the 'Delegation Map' for HITL/AITL across all business processes

The reality pointed out by Gartner that 'only 6% fully trust' demonstrates the structural fact that autonomy without the gradual building of trust increases organizational risk.

Using the three-mode classification presented in this whitepaper—HITL (final human judgment), AITL (autonomous agent processing), and HOTL (Human-On-The-Loop: higher-order monitoring via statistical quality control)—draw a 'Delegation Map' for all major business processes of your company. Article 14 of the EU AI Act (human oversight obligation) should function as its external compelling force.

### ③ Advance the introduction of Collaborative Robots (Cobots) and AMRs integrally with HITL design