Five archetypes: Simple Structure (direct supervision), Machine Bureaucracy (standardization of work processes), Professional Bureaucracy (standardization of skills), Divisionalized Form (standardization of outputs), Adhocracy (mutual adjustment). Six components: strategic apex, middle line, operating core, support staff, technostructure, and ideology (the sixth component added in Mintzberg 1989). Every coordination mechanism implicitly assumes that the actors being coordinated share human cognitive properties — understanding context, exercising judgment, bearing accountability. When actors include AI systems that can produce output without understanding, judgment, or accountability, the coordination mechanisms need rethinking, not just extension.

Four employment modes: (1) knowledge-based employment (high value, high uniqueness → commitment-based HR); (2) job-based employment (high value, low uniqueness → productivity-based HR); (3) contract work (low value, low uniqueness → compliance-based HR); (4) alliance/partnership (low value, high uniqueness → collaborative HR). AI actors break this framework: an AI system can be simultaneously high-value and zero-uniqueness (easily replicated), high-strategic-importance and zero-commitment (no employment relationship). The two-dimensional matrix (value × uniqueness) needs at minimum a third dimension: epistemic type (human assertion vs. AI inference).

The resource-based view treats human capital as a strategic resource (valuable, rare, inimitable, non-substitutable per Barney 1991). But AI actors violate multiple VRIO criteria: they are infinitely replicable (not rare), easily substitutable (swap one model for another), and their "capital" appreciates through version upgrades rather than experience accumulation. Agency theory assumes information asymmetry between principal and agent — but the asymmetry with AI agents is qualitatively different (the AI has no "interests" to misalign, but also no judgment to apply). Institutional theory (DiMaggio & Powell 1983) explains why organizations adopt similar HR practices through mimetic, coercive, and normative pressures — AI workforce governance is in the mimetic phase where organizations copy each other's ad-hoc approaches without theoretical grounding.

GPT-4 access boosted speed by 25%, human-rated performance by 40%, and task completion by 12% for tasks inside the frontier. For tasks outside the frontier, AI users performed 19 percentage points worse than non-users. The "jagged frontier" — the irregular boundary between tasks AI can and cannot do — is task-specific and not predictable from task difficulty. This finding has direct governance implications: policy cannot reliably specify which tasks to delegate to AI because the boundary is jagged, task-specific, and shifts with each model version. Governance infrastructure must allow dynamic, evidence-based assessment of AI capability per task scope — not static policy categorization.

AI governance requires a shift "from regulating what systems say to regulating how generative outputs are introduced into epistemic workflows, and where they may permissibly substitute for human judgment." This is precisely the epistemic break thesis: the problem is not that AI output is wrong — it is that AI output cannot be epistemically distinguished from human output without structural enforcement. Automation of epistemic authority fosters "epistemic passivity" — individuals relying on algorithmic assessments as proxies for judgment, weakening their own critical reasoning.

The paradox: organizations are advised to prioritize augmentation over automation for superior performance, but the boundary between augmentation and automation is unstable. A task that starts as human-with-AI-assistance can drift toward AI-with-human-oversight without anyone deciding to make the transition. This drift is ungoverneable through policy because policy assumes stable task categorization. Governance infrastructure must detect and manage the drift — tracking how the balance of human vs. AI contribution evolves for each scope over time.

Automation substitutes for labor on routine tasks (both manual and cognitive) while complementing labor on non-routine tasks requiring judgment, creativity, and interpersonal interaction. The complementarity effect raises wages for non-routine human tasks — the tasks that remain human become more valuable. For workforce governance: the twelve HR functions must be analyzed task-by-task (not function-by-function) to determine which sub-tasks are substituted, which are complemented, and which face the jagged frontier problem (F4). Performance management cannot be applied uniformly to a function that is 60% automated and 40% augmented.

Pattern A — Replaced by superior governance mechanism: Performance management → evidence-based graduation (continuous, per-scope, automatic regression). Compliance → structural enforcement (built into the record, not checked after the fact). Records management → typed decision lineage (the record IS the governance, not a separate activity).

Pattern B — Split into multiple AI-native categories: Recruitment splits into human recruitment + AI actor evaluation (model assessment against scope requirements) + hybrid team composition. Training splits into human development + AI configuration + world model orientation (different processes for different actor types).

Pattern C — New triggers and modalities: Offboarding gains a new modality: model migration (the "employee" gets a brain upgrade overnight — continuity and regression testing needed). Compensation gains AI cost accounting (token-based, usage-priced, fundamentally different from salary).

Pattern D — Unprecedented problems: Model version succession (no traditional analog), capacity planning across human/computational actors (different resource units), epistemic provenance tracking (which outputs were human-authored vs. AI-generated).

Team emergent states include team efficacy, empowerment, climate, cohesion, trust, collective cognition, and shared mental models. These states are social-psychological phenomena that arise from human interaction. When team members include AI actors that cannot experience trust, share mental models, or develop team cohesion, the IMOI framework's mediating variables need fundamental revision. AI actors can simulate the outputs of these states (producing consistent, reliable contributions) but cannot experience the states themselves. Performance assessment must distinguish between process quality (the mediating states) and output quality — and must account for the fact that AI actors have only the latter.

The c factor is predicted by social dynamics, not individual capability. AI actors lack social sensitivity and do not participate in conversational turn-taking in the same way humans do. This suggests that adding AI actors to a team does not simply add capability — it changes the social dynamics that produce collective intelligence. Performance governance must therefore assess not just individual contributions but the team's collective intelligence properties, and must track how those properties change as the human/AI composition shifts.

Hackman (2002) identified enabling conditions for effective teams: compelling direction, enabling structure, supportive context, expert coaching. These conditions are designed for human teams. For human-AI teams, "expert coaching" of an AI actor means configuration and prompt engineering; "enabling structure" includes governance infrastructure that types output by actor type; "compelling direction" must be computationally representable for AI actors to follow. Evidence-based graduation — where trust level adjusts per scope based on accumulated performance evidence, with automatic regression when performance degrades — is consistent with the academic trajectory toward continuous, evidence-based assessment. But no existing framework applies this specifically to mixed human-AI team performance.

When a model upgrades from v4 to v5, every certification, trust level, and performance record from v4 is potentially invalidated. The "employee" that earned trust on specific scopes may have different capabilities after the upgrade. Traditional succession planning assumes the predecessor and successor are different people; model migration is the same "actor" with different capabilities. Governance must address: when does a version upgrade require recertification? How much of the accumulated trust transfers? What regression testing is required?

Traditional workforce planning asks: "How many people with what skills do we need?" AI-native planning asks: "What is the right mix of human judgment-hours and computational inference-tokens for this scope?" The resource units are incommensurable without a governance framework that translates between them. Cost accounting is similarly mismatched: salaried humans have fixed costs with variable output; token-priced AI has variable costs with (within the frontier) predictable output.

Kellogg: "The hardest work isn't in deploying the model or writing smarter algorithms, but transforming the organization to support these things." Five heavy lifts identified: (1) data integration, (2) model validation including audit logs, (3) establishing economic value, (4) monitoring for drift, (5) governance — clarifying risks, security, legality, accountability at every step. The "for every hour perfecting a model, expect roughly four hours making it work in the real world" ratio directly supports the thesis that governance infrastructure is the primary workforce challenge, not technical capability.

SHRM theory (Wright & McMahan 1992, F3) theorizes human capital as a strategic resource — but does not extend to non-human capital with human-like output capabilities. AI governance literature (NIST AI RMF, EU AI Act, Jobin et al. 2019 from S2) theorizes responsible AI deployment — but does not ground this in workforce management theory. The intersection — "how do you manage a workforce that includes AI actors using governance infrastructure rather than policy?" — is the gap this sprint identifies.

Gartner predicts 20% of organizations will use AI to flatten organizational structures by 2026, eliminating half of middle management positions. WEF frames the transformation as "no longer about choosing between people and technology, but about designing systems." Industry sources converge on: (a) blended workforce models combining employees, partners, and AI agents; (b) governance requirements comparable to financial reporting or cybersecurity; (c) transparency about how AI influences hiring, promotion, and termination decisions; (d) credible proof of governance actions required by regulators, boards, and enterprise customers. All converge on the need — none provides the systematic framework.

NIST AI RMF, a voluntary framework, recommends: documented roles and responsibilities for AI risk management, organizational accountability structures, human oversight for high-impact systems, and audit trails. These are workforce governance functions — but the framework treats them as "AI governance" separate from "workforce management." The parallel-stack problem: organizations end up with one governance system for human workers (HR) and another for AI systems (AI governance), with no integration point. When a human and an AI actor collaborate on the same deliverable, which governance stack applies?