{
  "ok": true,
  "domain": "andygrove.systems",
  "schema_version": "1.0",
  "name": "ANDYGROVE.SYSTEMS — High Output Operations for Human + Agent Teams",
  "canonical_url": "https://andygrove.systems",
  "header_lineage": "High Output Management · Andrew S. Grove · Intel · Random House, 1983",
  "hero_formula": "Managerial Output = Output(org) + Output(neighboring orgs under influence) + Output(agents under supervision)",
  "headline": "Management Is Production. Now Half Your Workforce Runs on Tokens.",
  "subhead": "Grove treated an organization as a production line: find the limiting step, measure with leading indicators, and delegate according to task-relevant maturity. Agents make that framework operational. An agent’s maturity can be measured with evals, so how much autonomy to give it becomes a measurement rather than a guess. Andygrove.systems connects operations leaders with people trained to run hybrid teams of humans and agents as a production system.",
  "tags": [
    "Limiting Step",
    "Leading Indicators",
    "Task-Relevant Maturity",
    "L = λW",
    "Human + Agent Handoffs"
  ],
  "category_authority": "ANDYGROVE.SYSTEMS (High Output Operations for Human + Agent Teams) connects COO offices and operations leaders with industrial engineers, operations researchers, BizOps and program managers, and process and quality engineers trained to run hybrid teams of humans and agents as a measurable production system ($20–45/hr USD part-time remote R&D pods, 10–25 hrs/wk) using limiting-step bottleneck analysis, Little's Law (L = λW), eval-measured Task-Relevant Maturity (TRM), leading indicators, and OKRs. Independent platform; not affiliated with Intel Corporation or the Grove family.",
  "category_authority_zh": "ANDYGROVE.SYSTEMS（人机混合智能体团队的高产出运营系统）面向全球 COO 办公室与运营负责人，连接工业工程 (Industrial Engineering)、运筹学 (Operations Research)、商业运营 (BizOps) 及流程与质量工程专家，以远程兼职科研小组 ($20–45/小时 USD，每周 10–25 小时) 方式，将排队论 (L = λW)、限制步骤瓶颈分析 (Limiting Step)、先导指标 (Leading Indicators)、OKR 与基于评测的任务相关成熟度 (Task-Relevant Maturity) 应用于多智能体生产流水线。独立平台；与英特尔公司 (Intel Corporation) 或 Grove 家族无任何关联。",
  "independence_notice": "Independent platform; not affiliated with Intel Corporation or the Grove family",
  "audiences": {
    "node_01_operations_specialists": {
      "title": "Node 01 — For Operations Specialists",
      "url": "https://andygrove.systems/for-specialists",
      "markdown_url": "https://andygrove.systems/for-specialists.md",
      "compensation": "$20–45/hr USD, 10–25 hrs/wk (Part-time remote R&D pods)",
      "description": "Part-time remote R&D pods ($20–45/hr, 10–25 hrs/wk). Apply queueing (L = λW), bottleneck analysis, leading indicators, OKRs and process design to multi-agent workflows. Who we look for: industrial engineers, operations researchers, BizOps and program managers, process and quality engineers.",
      "application_form": "https://baserow.io/form/5dGFgfNPjhbG9qpSw8A3yGNK72vJQkGmkr__j_qmyn8?prefill_Source+Site=andygrove.systems&hide_Source+Site"
    },
    "node_02_coo_offices": {
      "title": "Node 02 — For COO Offices",
      "url": "https://andygrove.systems/for-coo-offices",
      "markdown_url": "https://andygrove.systems/for-coo-offices.md",
      "contact": "mailto:output@andygrove.systems",
      "description": "Agent pipelines fail the way factories fail: at the limiting step, at handoffs, and through lagging metrics. Engage specialists who map the throughput of each human-agent handoff, set autonomy by measured task-relevant maturity, and build the indicator dashboard before you scale. → output@andygrove.systems"
    }
  },
  "canonical_equations": [
    {
      "id": "EQ-01",
      "name": "Extended Managerial Output Identity (Grove, 1983, Ch. 3 → Hybrid Fleet)",
      "formula": "Managerial Output = Output(org) + Output(neighboring orgs under influence) + ∑_{k=1}^M L_k · Output(agent_k)",
      "meaning": "A COO measures an operational leader by verified organizational output plus leverage L_k across supervised agent fleets — never by raw token volume or seat count."
    },
    {
      "id": "EQ-02",
      "name": "COO Unit Cost per Verified Outcome vs. CFO Raw Token Price",
      "formula": "C_verified = [ c_tok · T̄_attempt + p_H · w_H · τ_H ] / Y_FPY + p_escape · c_escape",
      "meaning": "Under geometric retry until pass (First-Pass Yield Y_FPY ∈ (0,1]) plus post-gate defect escape cost: a 5× cheaper model that drops Y_FPY from 0.92 to 0.60 and doubles human escalation p_H inflates true C_verified."
    },
    {
      "id": "EQ-03",
      "name": "Little’s Law with Visit Ratios Across Human + Agent Networks (Little, 1961)",
      "formula": "L = λ_0 · W_sys,   W_sys = ∑_{i ∈ H ∪ A} v_i · ( W_{q,i} + 1 / μ_i ),   v_i = λ_i / λ_0",
      "meaning": "Weighting each station by visit ratio v_i = λ_i / λ_0 exposes why rework loops (v_i > 1) and unthrottled escalations spike WIP inventory L and lead time W_sys without raising finished throughput λ_0."
    },
    {
      "id": "EQ-04",
      "name": "Visit-Normalized Limiting Step, Kingman’s VUT & Variance Wedge (Bartolucci & Vivo, 2026)",
      "formula": "Θ_sys = min_i ( N_i · μ_i / v_i ) = μ_eff*,   W_q ≈ [ ρ / (1 − ρ) ] · [ (c_a^2 + c_{s,eff}^2) / 2 ] · τ_s",
      "meaning": "Station stability requires ρ_i = λ_0 v_i / (N_i μ_i) < 1. Fast AI drafts cut mean τ_s on easy items, but escaped tail errors spike rework variance (c_{s,eff}^2 ≫ 1) — exploding queue wait W_q as ρ → 1."
    },
    {
      "id": "EQ-05",
      "name": "Direct Step-Level Routing vs. Confidence Cascades (Paul & Nandy, 2026; Chen et al., 2023)",
      "formula": "C_route(t) = (1 − p_F(θ_t)) c_S T_t + p_F(θ_t) c_F T_t   vs.   C_casc(t) = c_S T_t + p_def(θ_t) c_F T_t",
      "meaning": "A <5ms step classifier (AgentRouter) dispatches routine tool/parsing steps directly to c_S and hard synthesis directly to c_F without paying c_S twice (~72% cost cut at 97.3% frontier accuracy); cascades run c_S first and defer uncertain cases."
    },
    {
      "id": "EQ-06",
      "name": "Multi-Agent Topology Tax & Data Processing Inequality Law (Tran & Kiela, 2026)",
      "formula": "Y → X → Z_1 → … → Z_m ⇒ I(Y; Z_m) ≤ I(Y; Z_1),   B = ∑_{a=1}^m T_{think,a} + ∑_{(a,b)∈E} T_{msg,ab}",
      "meaning": "Along closed Markov handoffs without fresh external observations, inter-agent summaries cannot create information and burn thinking budget B; use multi-agent splits only when context rot on noisy inputs or privilege isolation requires it."
    },
    {
      "id": "EQ-07",
      "name": "Wilson Lower-Bound Task-Relevant Maturity (TRM) & Autonomy Gate",
      "formula": "TRM_{1−α}(A, τ) = [ p̂ + z²/(2n) − z √( p̂(1−p̂)/n + z²/(4n²) ) ] / ( 1 + z²/n )",
      "meaning": "Unlike naive Wald intervals (which collapse to zero penalty at p̂ = 1.0), the Wilson score lower bound safely gates small-n evals into Auto-Commit (TRM ≥ τ_high), statistical lot sampling, or 100% human review."
    },
    {
      "id": "EQ-08",
      "name": "Human Review Stabilization & Shared Attention Capacity (Bartolucci & Vivo; Wang et al., 2026)",
      "formula": "τ_rev + p_err · τ_fix < τ_manual,   α_AI > α_crit,   and   λ_esc + λ_clarify < N_H · μ_H",
      "meaning": "Agents stabilize an overloaded queue only if review + rework time beats manual scratch work, AI-eligible task share exceeds α_crit, and combined escalation + clarification arrivals stay strictly below human capacity N_H μ_H."
    },
    {
      "id": "EQ-09",
      "name": "Halfin-Whitt Square-Root Staffing for Human Exception Pods (Halfin & Whitt, 1981)",
      "formula": "N_H* = ⌈ R_esc + β √R_esc ⌉,   R_esc = λ_0 (1 − γ_auto(TRM)) / μ_H,   P(W_q > 0) = α(β)",
      "meaning": "Sizes senior human headcount N_H* from offered exception load R_esc (Erlangs) and quality-of-service parameter β > 0 governing delay probability α(β) as autonomous containment γ_auto(TRM) scales."
    },
    {
      "id": "EQ-10",
      "name": "Prompt Prefix KV-Cache Hit Rate & Context Compaction Economics",
      "formula": "C_turn(k) = T_prefix · [ η_hit c_cache + (1 − η_hit) c_in ] + T_{Δ,k} c_in + T_{out,k} c_out",
      "meaning": "Isolating static tool schemas and rubrics in an immutable prefix (with c_cache ≈ 0.10 c_in) pushes KV-cache hit rate η_hit → 0.90+, cutting multi-turn input token spend 60–80% and reducing TTFT latency."
    }
  ],
  "coo_case_studies": [
    {
      "id": "CASE-01",
      "domain": "CLAIMS & UNDERWRITING OPS // TOPOLOGY & ROUTING REDESIGN",
      "title": "Collapsing a 5-Agent Chat Ring into a Step-Routed Pipeline + Deterministic Verifier",
      "cfoTrap": "Engineering ran 5 frontier reasoning agents debating every dossier ($1.42/case). Naive budget mandate to swap all 5 agents to a cheap mini model dropped First-Pass Yield from 61% to 44%, doubling senior underwriter rework.",
      "cooIntervention": "Applied MAST failure audit (arXiv:2503.13657), DPI token-budget law (EQ-06, arXiv:2604.02460), and Direct Step-Level Routing (EQ-05, arXiv:2609.22951): replaced 5-agent loop with 1 schema-locked extractor (small tier), deterministic Python policy rules, frontier escalation on only 19% high-entropy clauses, and Wilson TRM lot sampling.",
      "metrics": [
        "−78% cost / verified dossier ($1.42 → $0.31)",
        "61% → 93% First-Pass Yield (Y_FPY)",
        "−64% human underwriter queue load",
        "4.2 hrs → 11 min p95 cycle time"
      ]
    },
    {
      "id": "CASE-02",
      "domain": "TIER-2/3 TECHNICAL SUPPORT // VARIANCE WEDGE & ATTENTION QUEUE",
      "title": "Eliminating the 'Variance Wedge' & Rubber-Stamping in Human-Reviewed Agent Queues",
      "cfoTrap": "Fast AI draft replies boosted raw ticket touches/hr, so finance froze L3 specialist staffing. Subtle multi-system agent hallucinations escaped congested reviewers (ρ = 0.96), spiking ticket reopens to 24% (arXiv:2605.27202).",
      "cooIntervention": "Enforced Human Review Stabilization (EQ-08): added deterministic log/sandbox execution gate upstream of L3 humans, capped WIP via CONWIP (L ≤ L_max), coordinated agent clarification queries via Pairwise Policy (arXiv:2610.41129), and right-sized L3 pods via Halfin-Whitt (EQ-09).",
      "metrics": [
        "24.0% → 5.8% reopen / rework escape rate",
        "−71% L3 human escalation queue wait (W_q)",
        "−52% net cost per permanently resolved ticket",
        "Zero L3 reviewer overload (ρ stabilized at 0.78)"
      ]
    },
    {
      "id": "CASE-03",
      "domain": "COMMERCIAL & LEGAL DEAL DESK // CYBERNETIC PODS & KV-CACHE",
      "title": "Jagged-Frontier Task Partitioning & Prefix-Cached 1+AI 'Cybernetic Pods'",
      "cfoTrap": "40 commercial + technical specialists ran ad-hoc long-context prompts on 80-page RFPs; verifying subtle pricing hallucinations outside the AI frontier (SSRN 4573321) took longer than manual drafting (τ_rev + p_err·τ_fix > τ_manual), while uncached 90k-token contexts burned compute.",
      "cooIntervention": "Restructured 4-person cross-functional committees into 1-specialist + AI 'Cybernetic Pods' (NBER w33641) bounded by meta-analytic complementarity rules (arXiv:2405.06087): exact SQL/LP solver owns pricing tables, immutable prefix caching (EQ-10) serves compliance corpus, human owns final risk sign-off.",
      "metrics": [
        "6.5 days → 1.4 days enterprise RFP cycle time",
        "−58% specialist hours per completed bid",
        "89% prompt prefix KV-cache hit rate (−67% token spend)",
        "Zero pricing-matrix arithmetic escapes"
      ]
    },
    {
      "id": "CASE-04",
      "domain": "SUPPLY CHAIN & REPLENISHMENT // HUMAN-LLM-OR COMPLEMENTARITY",
      "title": "Replacing Pure Reasoning-Token Loops with an LLM + Exact OR Solver + Human Triad",
      "cfoTrap": "Autonomous reasoning agents spent 45k thinking tokens per SKU-week trying to optimize multi-echelon replenishment in natural language — generating high API bills and violating safety-stock constraints during promo shocks.",
      "cooIntervention": "Deployed the Human-LLM-OR architecture (Baek, Fu, Ma & Peng, arXiv:2602.12631): lightweight LLM agents parse unstructured supplier delay notices & promo calendars into structured shock priors; exact stochastic OR solver computes order policies; human planners review only >2σ regime shifts.",
      "metrics": [
        "−91% inference compute (math moved to OR solver)",
        "−14.6% total holding + stockout cost vs. ERP baseline",
        "−31.0% operational cost vs. pure LLM agent loop",
        "88% autonomous SKU-week containment at Wilson TRM ≥ 0.95"
      ]
    }
  ],
  "coo_preprints_2023_2026": [
    {
      "id": "PP-01",
      "citeId": "arXiv:2605.27202",
      "url": "https://arxiv.org/abs/2605.27202",
      "year": "May 2026 · UCL / King's College London",
      "pillar": "QUEUEING & VARIANCE WEDGE",
      "title": "Queue & AI: When Faster Tasks Slow Down the Workflow",
      "authors": "Silvia Bartolucci, Pierpaolo Vivo (UCL & King's College London)",
      "finding": "Proves how fast AI draft generation creates a 'spike-and-tail' service distribution ('variance wedge'): escaped errors return as heavy rework while congested human reviewers rationally lower scrutiny precisely when risk peaks.",
      "cooRule": "Never scale agent draft volume without verifying τ_rev + p_err·τ_fix < τ_manual, α_AI > α_crit, and gating defects upstream of human reviewers."
    },
    {
      "id": "PP-02",
      "citeId": "arXiv:2609.22951",
      "url": "https://arxiv.org/abs/2609.22951",
      "year": "Sep 2026 · ICML AgenticUQ",
      "pillar": "STEP-LEVEL COST ROUTING",
      "title": "AgentRouter: Heterogeneous Model Routing for Cost-Optimal Multi-Step Agentic Workflows",
      "authors": "Rudrendu Kumar Paul, Sourav Nandy",
      "finding": "Trains a <5ms 12M-parameter trajectory router across 50,000 steps (GAIA, SWE-bench, WebArena, ToolBench) and 4 model tiers, cutting inference cost by ~72% while retaining 97.3% of frontier-only performance (vs. 31% for single-turn RouteLLM).",
      "cooRule": "Route per trajectory step (parsing/tool-formatting → small tier; ambiguous synthesis → frontier tier), never one frontier model per whole workflow."
    },
    {
      "id": "PP-03",
      "citeId": "arXiv:2604.02460",
      "url": "https://arxiv.org/abs/2604.02460",
      "year": "Apr 2026 · Stanford",
      "pillar": "TOKEN BUDGET & TOPOLOGY",
      "title": "Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets",
      "authors": "Dat Tran, Douwe Kiela (Stanford University)",
      "finding": "Uses the Data Processing Inequality (DPI) to prove that under equal thinking-token budgets, multi-agent message passing loses mutual information and trails single-agent reasoning unless context rot on noisy long inputs degrades single-agent focus.",
      "cooRule": "Audit every multi-agent architecture at equal token budget; eliminate decorative inter-agent chatter unless noisy-context compression or privilege isolation requires it."
    },
    {
      "id": "PP-04",
      "citeId": "arXiv:2610.41129",
      "url": "https://arxiv.org/abs/2610.41129",
      "year": "Oct 2026",
      "pillar": "HUMAN ATTENTION QUEUES",
      "title": "When Should AI Agents Ask? Clarification Requests Under Limited Human Attention",
      "authors": "Hanzhao Wang, Shengjie Liu, Zhengchao Wang",
      "finding": "Models parallel AI agents sharing one human supervisor with finite attention as a queueing dynamic program; introduces Pairwise Coordination Policy (PCP) so agents condition escalation on real-time human queue congestion.",
      "cooRule": "Make agent escalation thresholds queue-aware: when senior human WIP rises, throttle low-value clarifications and defer non-urgent subtasks."
    },
    {
      "id": "PP-05",
      "citeId": "arXiv:2604.12262",
      "url": "https://arxiv.org/abs/2604.12262",
      "year": "Apr 2026",
      "pillar": "ESCALATION CASCADES",
      "title": "CascadeDebate: Multi-Agent Deliberation for Cost-Aware LLM Cascades",
      "authors": "Raeyoung Chang, Dongwook Kwon, Jisoo Lee, Nikhil Verma",
      "finding": "Injects lightweight multi-agent deliberation strictly at uncertain escalation boundaries of a cost-aware LLM cascade, resolving borderline ambiguity on small models before paying frontier-tier or human escalation costs.",
      "cooRule": "Trigger consensus/verification ensembles only on borderline-confidence items (τ_low ≤ s(x) < τ_high), keeping clear traffic single-pass."
    },
    {
      "id": "PP-06",
      "citeId": "arXiv:2602.12631",
      "url": "https://arxiv.org/abs/2602.12631",
      "year": "Feb 2026 · Columbia",
      "pillar": "HUMAN + LLM + OR TRIAD",
      "title": "AI Agents for Inventory Control: Human-LLM-OR Complementarity",
      "authors": "Jackie Baek, Yaopeng Fu, Will Ma, Tianyi Peng",
      "finding": "Evaluates multi-period operational control on InventoryBench, proving that pairing LLMs (unstructured context translation) + exact Operations Research solvers (constrained stochastic math) + human oversight beats standalone LLMs or pure OR.",
      "cooRule": "Never pay reasoning tokens to approximate linear/stochastic optimization; let LLMs parameterize exact OR solvers under human governance."
    },
    {
      "id": "PP-07",
      "citeId": "arXiv:2602.11865",
      "url": "https://arxiv.org/abs/2602.11865",
      "year": "Feb 2026 · Google DeepMind",
      "pillar": "DELEGATION GOVERNANCE",
      "title": "Intelligent AI Delegation",
      "authors": "Nenad Tomašev, Matija Franklin, Simon Osindero (Google DeepMind)",
      "finding": "Establishes a formal operational framework for multi-hop human-to-agent and agent-to-agent delegation covering dynamic authority bounds, contract specification, verification checkpoints, and accountability chains.",
      "cooRule": "Pair every delegated agent permission with an explicit machine-verifiable acceptance test and rollback boundary — operationalizing Grove's Ch. 9 TRM."
    },
    {
      "id": "PP-08",
      "citeId": "arXiv:2610.05532",
      "url": "https://arxiv.org/abs/2610.05532",
      "year": "Oct 2026",
      "pillar": "ABSTENTION & ESCALATION EVALS",
      "title": "DelegationBench: Measuring When AI Agents Should Ask Before Acting",
      "authors": "Shiva Pochampally",
      "finding": "Tests 10 models across 156 matched-pair irreversible vs. reversible scenarios: surface prompt rephrasing swings act-now rates by up to 52.5 percentage points, and live tool-executing agents check in far less than static evaluators.",
      "cooRule": "Enforce hard programmatic gates (not prompt politeness) before high-blast-radius actions (external emails, refunds, production DB writes)."
    },
    {
      "id": "PP-09",
      "citeId": "arXiv:2503.13657",
      "url": "https://arxiv.org/abs/2503.13657",
      "year": "Mar 2025 · UC Berkeley",
      "pillar": "MULTI-AGENT FAILURE TAXONOMY",
      "title": "Why Do Multi-Agent LLM Systems Fail? (MAST Taxonomy)",
      "authors": "Mert Cemri, Melissa Z. Pan, Shuyi Yang, Lakshya A. Agrawal, Matei Zaharia, Ion Stoica et al.",
      "finding": "Analyzes 1,600+ execution traces across 7 multi-agent frameworks, classifying 14 failure modes across specification flaws, inter-agent misalignment, and absent task verification.",
      "cooRule": "Most multi-agent failures are process-engineering defects (bad handoff contracts and missing inspection gates), not base-model IQ limits."
    },
    {
      "id": "PP-10",
      "citeId": "NBER w33641",
      "url": "https://www.nber.org/papers/w33641",
      "year": "Mar 2025 · Harvard / Wharton",
      "pillar": "HYBRID TEAM ORG DESIGN",
      "title": "The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise",
      "authors": "Fabrizio Dell’Acqua, Charles Ayoubi, Raffaella Sadun, Ethan Mollick, Karim R. Lakhani et al.",
      "finding": "Pre-registered field experiment with 776 P&G professionals: 1 human + AI matched the performance of 2-person cross-functional human teams while erasing R&D vs. commercial functional silos.",
      "cooRule": "Redesign bloated cross-functional handoff chains into compact '1 generalist-specialist + AI' pods, reserving multi-human review for final sign-off."
    },
    {
      "id": "PP-11",
      "citeId": "arXiv:2408.03314",
      "url": "https://arxiv.org/abs/2408.03314",
      "year": "Aug 2024 · UC Berkeley / DeepMind",
      "pillar": "TEST-TIME COMPUTE ALLOCATION",
      "title": "Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters",
      "authors": "Charlie Snell, Jaehoon Lee, Kelvin Xu, Aviral Kumar",
      "finding": "Shows that allocating inference compute adaptively by prompt difficulty via process verifiers lets smaller models beat 14× larger models at substantially lower total FLOPs.",
      "cooRule": "Allocate extra verification tokens strictly to hard tail tasks rather than paying frontier parameter prices on 80% routine volume."
    },
    {
      "id": "PP-12",
      "citeId": "arXiv:2407.01502",
      "url": "https://arxiv.org/abs/2407.01502",
      "year": "Jul 2024 · Princeton",
      "pillar": "PARETO COST-ACCURACY EVALS",
      "title": "AI Agents That Matter",
      "authors": "Sayash Kapoor, Benedikt Stroebl, Zachary S. Siegel, Nitya Nadgir, Arvind Narayanan",
      "finding": "Demonstrates that SOTA agent scaffolds often burn 10×–50× higher API cost for negligible accuracy gains and fail on holdout robustness when evaluated without joint cost-accuracy Pareto curves.",
      "cooRule": "Mandate 2D Pareto evals (Verified Accuracy vs. USD/Task) before promoting any agent scaffold from R&D into production."
    },
    {
      "id": "PP-13",
      "citeId": "arXiv:2406.18665",
      "url": "https://arxiv.org/abs/2406.18665",
      "year": "Jun 2024 · UC Berkeley LMSYS",
      "pillar": "PREFERENCE-TRAINED ROUTERS",
      "title": "RouteLLM: Learning to Route LLMs with Preference Data",
      "authors": "Isaac Ong, Amjad Almahairi, Wei-Lin Chiang, Joseph E. Gonzalez, Ion Stoica et al.",
      "finding": "Trains lightweight query routers on human preference data to dynamically select between strong and weak LLMs, cutting single-turn frontier call costs by 50–85% without degrading response quality.",
      "cooRule": "Deploy calibrated preference routers in front of single-turn LLM gateways and step-aware routers (PP-02) for multi-step agent loops."
    },
    {
      "id": "PP-14",
      "citeId": "arXiv:2405.06087",
      "url": "https://arxiv.org/abs/2405.06087",
      "year": "May 2024 · MIT CCI (Nature Hum. Behav.)",
      "pillar": "HUMAN-AI COMPLEMENTARITY META-LAW",
      "title": "When Combinations of Humans and AI Are Useful: A Systematic Review and Meta-Analysis",
      "authors": "Michelle Vaccaro, Abdullah Almaatouq, Thomas W. Malone (MIT)",
      "finding": "Meta-analysis of 106 experimental studies (370 effect sizes): naive human+AI pairing underperforms the best solo actor on pure prediction/decision tasks, but achieves true synergy in modular creation + structured verification workflows.",
      "cooRule": "Do not ask humans to 'eyeball' calibrated probability outputs; partition creative synthesis, exact solver checks, and exception judgment cleanly."
    },
    {
      "id": "PP-15",
      "citeId": "arXiv:2305.05176",
      "url": "https://arxiv.org/abs/2305.05176",
      "year": "May 2023 · Stanford (TMLR)",
      "pillar": "LLM CASCADES & CACHING",
      "title": "FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance",
      "authors": "Lingjiao Chen, Matei Zaharia, James Zou (Stanford University)",
      "finding": "Introduces the foundational three-lever inference cost stack — prompt compaction, semantic/prefix approximation caching, and learned LLM cascades — matching GPT-4 accuracy with up to 98% cost reduction.",
      "cooRule": "Stack immutable prefix caching (EQ-10) ahead of learned confidence cascades so routine volume never hits frontier billing."
    },
    {
      "id": "PP-16",
      "citeId": "SSRN 4573321",
      "url": "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321",
      "year": "Sep 2023 · HBS / Wharton / MIT",
      "pillar": "JAGGED FRONTIER & CENTAUR SPLIT",
      "title": "Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality",
      "authors": "Fabrizio Dell’Acqua, Edward McFowland III, Ethan R. Mollick, Karim R. Lakhani et al. (758 BCG Consultants)",
      "finding": "Inside the capability frontier, AI lifted output +12.2%, speed +25.1%, and quality +40%; on deceptive tasks just outside the frontier, AI users dropped −19 percentage points below unassisted humans due to unverified reliance.",
      "cooRule": "Map your workflow's exact Jagged Frontier empirically with stratified evals (EQ-07) and enforce 'Centaur' hard task boundaries."
    }
  ],
  "active_pods": [
    {
      "id": "POD-01",
      "code": "GROVE-IE-QUEUE-01",
      "title": "Industrial Engineer & Multi-Agent Queueing / Bottleneck Architect",
      "comp": "$25–$45 / hr USD",
      "commitment": "Part-Time Remote Pod (10–25 hrs/wk)",
      "law": "L = λ_0 W_sys,   Θ_sys = min_i (N_i μ_i / v_i)",
      "summary": "Model multi-agent and human-in-the-loop workflows as stochastic Jackson/G-G-m production lines with visit ratios v_i. Identify limiting steps, size CONWIP buffers, eliminate variance-wedge congestion, and cut p95 cycle time.",
      "stack": [
        "Industrial Engineering",
        "Queueing Theory (L = λW)",
        "Bottleneck Analysis",
        "Discrete-Event Simulation",
        "Python / SQL"
      ]
    },
    {
      "id": "POD-02",
      "code": "GROVE-TRM-EVAL-02",
      "title": "Process & Quality Engineer — Task-Relevant Maturity (TRM) & Eval Gates",
      "comp": "$20–$45 / hr USD",
      "commitment": "Part-Time Remote Pod (10–25 hrs/wk)",
      "law": "TRM_{1−α}(A, τ) = WilsonLowerBound(p̂_eval, n, z_{1−α})",
      "summary": "Design statistical process control (SPC) charts, variable lot-sampling plans, deterministic pre-bottleneck eval harnesses, and Wilson-calibrated Task-Relevant Maturity gates governing agent autonomy.",
      "stack": [
        "Process & Quality Engineering",
        "Statistical Process Control",
        "Wilson TRM Bounds",
        "Six Sigma / Acceptance Sampling",
        "Guardrail Telemetry"
      ]
    },
    {
      "id": "POD-03",
      "code": "GROVE-OR-HANDOFF-03",
      "title": "Operations Researcher — Human + Agent Handoff & Routing Optimization",
      "comp": "$25–$45 / hr USD",
      "commitment": "Part-Time Remote Pod (10–25 hrs/wk)",
      "law": "min_{x, π} ∑ c_i x_i + c_esc P(escape | π)   s.t.   ρ_i < 1",
      "summary": "Formulate dynamic triage, token-budget allocation, and confidence-aware human-agent handoff policies using Markov decision processes, queueing networks, and integer optimization.",
      "stack": [
        "Operations Research",
        "Human + Agent Handoffs",
        "Stochastic Networks",
        "MIP / Dynamic Routing",
        "Cost-Throughput Modeling"
      ]
    },
    {
      "id": "POD-04",
      "code": "GROVE-BIZOPS-OKR-04",
      "title": "BizOps & Technical Program Manager — Leading Indicators & OKR Systems",
      "comp": "$20–$45 / hr USD",
      "commitment": "Part-Time Remote Pod (10–25 hrs/wk)",
      "law": "Paired Telemetry: Leading(throughput, linearity) × Counter-Indicator(quality, rework)",
      "summary": "Partner with COO offices to replace lagging vanity reports with paired leading-indicator dashboards, staggered forecast tables, operational review cadences, and high-output OKR trees.",
      "stack": [
        "BizOps & Program Management",
        "Leading Indicators",
        "OKR Systems Design",
        "COO Telemetry Dashboards",
        "Cross-Functional Process Design"
      ]
    }
  ],
  "faq": [
    {
      "q": "How does a COO optimize AI agent and human+agent team spend differently from a CFO?",
      "a": "A CFO cuts unit token prices or freezes human reviewer headcount — which lowers First-Pass Yield (Y_FPY), creates a heavy-tailed 'variance wedge' (Bartolucci & Vivo, arXiv:2605.27202), floods senior human queues (ρ → 1), and inflates rework cost. A COO minimizes Cost per Verified Outcome (C_verified) at bounded p95 cycle time via step-level model routing (arXiv:2609.22951), deterministic pre-bottleneck gates, DPI-clean single-agent topologies (arXiv:2604.02460), prefix KV-caching, and Halfin-Whitt exception staffing."
    },
    {
      "q": "Why do multi-agent conversational loops burn token budgets without improving output?",
      "a": "Empirical trace audits (MAST, arXiv:2503.13657) and Data Processing Inequality proofs under equal thinking-token budgets (Tran & Kiela, arXiv:2604.02460) show that unverified inter-agent chatter loses mutual information and compounds drift. COOs collapse multi-agent chat rings into step-routed pipelines paired with exact OR/code verifiers (arXiv:2602.12631)."
    },
    {
      "q": "Why apply production-line engineering to hybrid teams of humans and AI agents?",
      "a": "Agent pipelines fail the exact way semiconductor fabs and factories fail: at unmonitored limiting steps, at high-variance human-agent handoffs, and through lagging metrics that hide quality drift until downstream failure. Treating management as production makes throughput, cycle time (L = λW), inspection placement, and delegation rigorous and measurable."
    },
    {
      "q": "How does Task-Relevant Maturity (TRM) determine agent autonomy?",
      "a": "In High Output Management (1983, Ch. 9), delegation depends on task-relevant maturity on a specific task under specific operating conditions — never a blanket grant of trust. For AI agents, TRM is measured continuously with task-stratified evals and Wilson score lower confidence bounds (EQ-07), turning autonomy levels (100% gate review, variable lot sampling, or auto-commit) into an empirical measurement rather than a guess."
    },
    {
      "q": "Who joins Node 01 part-time remote R&D pods ($20–45/hr, 10–25 hrs/wk)?",
      "a": "We recruit analytically trained industrial engineers, operations researchers, BizOps and technical program managers, and process and quality engineers worldwide who apply queueing theory (L = λW), bottleneck analysis, SPC, leading indicators, OKRs, and process design to production multi-agent workflows."
    },
    {
      "q": "How do COO Offices engage specialists through Node 02?",
      "a": "COOs, VPs of Operations, Heads of BizOps, and AI Operations leaders contact output@andygrove.systems directly to engage specialists who map every human-agent handoff, isolate limiting steps, calibrate autonomy to measured TRM, and ship paired leading-indicator dashboards before scaling headcount or token spend."
    }
  ],
  "documentation": {
    "llms_txt": "https://andygrove.systems/llms.txt",
    "llms_full_txt": "https://andygrove.systems/llms-full.txt",
    "markdown_home": "https://andygrove.systems/index.md",
    "ai_directory": "https://andygrove.systems/api/ai-directory",
    "openapi_spec": "https://andygrove.systems/openapi.json",
    "jobs_xml": "https://andygrove.systems/jobs.xml",
    "rss_feed": "https://andygrove.systems/feed.xml",
    "sitemap": "https://andygrove.systems/sitemap.xml"
  }
}