Rewired (2nd ed., Lamarre, Smaje, Levin 2024): Six-Capability Synthesis

Brandon Sneider · April 2026

Source: [sources/04-consulting-firms/mckinsey-rewired-2nd-edition-raw.md](../../sources/04-consulting-firms/mckinsey-rewired-2nd-edition-raw.md)

Source: sources/04-consulting-firms/mckinsey-rewired-2nd-edition-raw.md

Executive Summary

Rewired’s central claim — that enterprise AI value is gated by six mutually reinforcing capabilities, not by model choice or tool selection — aligns with the independent corpus evidence (BCG 10/20/70, MIT CISR n=721, Stanford HAI 2026) that 70–90% of AI value resides in people, operating model, and workflow design.
The book’s six capabilities (roadmap, talent, operating model, technology, data, adoption) are best read as a forcing function for CEO leadership, not as a maturity ladder. The 2nd edition’s additions (Ch. 5 agentic workflows, Ch. 11 agentic talent model, Ch. 22 agentic engineering) tighten the playbook for 2026 conditions.
Two tensions to flag when using Rewired with Fortune 500 clients: (1) the book’s value-capture confidence runs ahead of independent field evidence (NBER 2026 found 89% of firms report zero productivity impact); (2) the “build it yourself” stance on AI platforms and agentic engineering understates the realistic mid-market path, which is usually SI + platform.
For CIOs and transformation leaders, the most actionable pieces are Ch. 3 (domains over use cases), Ch. 5 (agentic workflow selection criteria), Ch. 25 (no data architecture, no AI advantage), and Ch. 30 (adoption stick). These four chapters resolve the bulk of the pilot-to-production gap documented elsewhere in this corpus.

Capability 1 — Bold strategy and roadmap (Ch. 1–7)

Focus on business domains, not use cases (Ch. 3, p.49). Use cases scattered across a hundred initiatives produce the “use-case implementation race” anti-pattern (Introduction pp.20–21) — high activity, no P&L impact. Domains (customer acquisition, claims, supply planning) are large enough to re-sequence end-to-end and small enough to own.
Business leaders lead the reimagination (Ch. 4, p.63). The domain owner is a business leader, not the CIO. Air Canada is the referenced example. This inverts the 65–70% of enterprises where the CIO or a CAIO is handed the transformation brief.
Economics of the transformation (Ch. 2, p.39): the book argues that successful transformations return “substantial” economic value but does not disclose distributional statistics (top-quartile vs. median vs. bottom-quartile). Cross-reference: Grant Thornton and PwC independent evidence shows the median firm captures modest returns; value concentration at the top decile is well-documented (see wiki/ai-roi-evidence.md).

Capability 2 — Talent bench (Ch. 8–11)

Business-leader tech fluency is non-negotiable (Ch. 8, p.131). The book treats business-leader tech literacy as a skill build, not a delegation target. This aligns with the Spencer Stuart / Stanford board-AI-literacy finding that <20% of Fortune 500 boards have two or more directors with operating AI experience.
IT talent “surgery” (Ch. 9, p.141): the expanded chapter frames the IT talent reshape as triage — retain the top builders, remove drag, recruit missing roles (ML engineer, data engineer, platform engineer, SRE). This is incompatible with the “fully outsourced IT” posture documented in research/07-adoption-challenges/ai-adoption-fully-outsourced-it.md.
Agentic-era talent model (Ch. 11, p.175): the book anticipates role evolution toward workflow designer, agent supervisor, and eval engineer. This is consistent with BCG’s “structure to flow” findings and Google Cloud’s digital-assembly-line framing.

Capability 3 — Operating model (product/agile) (Ch. 12–16)

Pod-based operating model (Ch. 12, p.199) with embedded engineering (Ch. 13, p.217). The prescription is stable, business-aligned, cross-functional pods that own a product or workflow end-to-end. This matches Stanford HAI 2026’s finding that deployments with human-primary architectures capture a fraction of the value that pod-owned, AI-primary workflows capture.
Blueprint for a faster enterprise (Ch. 14, p.237) and controlled transition (Ch. 15, p.257). The book acknowledges explicitly that most enterprises must transition existing units rather than greenfield — Ch. 16 (p.265) covers the greenfield case separately.

Capability 4 — Technology and data backbone (Ch. 17–29)

“No data architecture, no AI advantage” (Ch. 25, p.391). The book’s strongest stance: without a governed, reusable data architecture, AI investments produce recurring bespoke-integration costs and bounded outcomes. This is directly consistent with Palantir’s ontology-first commercial argument and with Bain’s Phase 1 (Foundation) in the agentic deployment sequence.
Data products as reusable building blocks (Ch. 26, p.401). Each high-value data product (e.g., customer 360, supplier-risk profile) is built once and reused across many AI workloads. This is the economic mechanism behind the book’s “reuse case” thesis.
AI platform to accelerate model development (Ch. 21, p.325) and agentic engineering platform (Ch. 22, p.337). The book advocates a build-side platform investment. Mid-market adaptation: for most enterprises below the Fortune 500 band, the equivalent is purchasing a platform (Snowflake + Cortex / Databricks / Vertex / Foundry) rather than building one.
Protecting data in an LLM world (Ch. 28, p.427). Covers data leakage, model-memorization risk, and the operational controls needed when enterprise data is exposed to LLMs.

Capability 5 — Adoption and scaling (Ch. 30–33)

Make adoption stick (Ch. 30, p.457). The book’s explicit position: adoption is engineered, not wished for. Treating agents like new employees — define the job, onboard carefully, evaluate continuously — is the paraphrasable principle (Ch. 5, p.92).
Design for scale and reuse from day one (Ch. 31, p.469). “The best use case is the reuse case” (Ch. 5, p.93; Ch. 31). This directly opposes the “scattered pilots” pattern documented in McKinsey’s own state-of-AI surveys and in BCG AI Radar.
Track what matters (Ch. 32, p.483). KPI discipline — impact tied to P&L line items, not activity counts.
Plan for midstream adjustments (Ch. 33, p.501). Explicit acknowledgment that the first plan will be wrong; the operating discipline is re-planning, not plan perfection.

Capability 6 — Governance and risk management (Ch. 34–35)

Risk, trust, and the right to deploy AI (Ch. 34, p.511). The expanded chapter argues for a “right-to-deploy” gating framework — legal, risk, compliance, and security sign-off as a parallel stream to engineering. This is closer to Bain’s Phase 1 (Foundation & Governance) than to a “move fast and fix later” posture.
Culture as an enabler (Ch. 35, p.523). The book’s closing argument is that culture — psychological safety for experimentation, tolerance for midstream adjustments, CEO-visible learning — is the multiplier that makes the other five capabilities compound.

Agentic AI integration (new in 2nd ed.)

Four levels of agentic automation (Ch. 5, p.82, Exhibit 5.1): (1) individual augmentation, (2) task automation, (3) agentic workflows, (4) agentic systems. Workflow selection criteria: dynamic, unstructured, edge-case-heavy work benefits from agents; deterministic, structured, high-volume work benefits from rules-based systems. “Choosing the right tool for the job is not a sign of technological conservatism; it’s a mark of maturity” (Ch. 5, p.92).
Evaluation discipline: task success rate, F1/precision/recall, retrieval accuracy, hallucination rate (Ch. 5, p.94 sidebar). Paired with continuous frontline involvement — not a one-time eval built at deployment.
Case evidence: E.ON Next (Ch. 5, pp.79–80) — +6pp CSAT, −8% AHT, +14% transaction success, ~50% cost-per-call reduction. The book’s one named, measured agentic deployment.

Key tensions with other corpus findings

Tension 1 — Value capture confidence vs. NBER/METR/MIT CISR evidence

Rewired’s framing — “companies that succeed deliver substantial economic returns” (Introduction; Ch. 2) — is directionally consistent with the top-quartile outcome, but the book does not bracket the distribution. The independent evidence does:

NBER (Feb 2026, n=5,867 executives across 4 countries): 89% of firms report zero productivity impact from AI over the preceding three years.
METR (Jul 2025 RCT, field study): experienced developers 19% slower with AI tools while believing they were 20% faster.
MIT CISR (n=721): Stage 1 (tools without workflow redesign) = −12.6 pp growth vs. industry average; Stage 3+ (with redesign) = +11.3 to +17.1 pp.
Faros AI (10,000+ devs): 98% more PRs after Copilot rollout; zero delivery throughput improvement.

The synthesis: Rewired’s capabilities are necessary but not sufficient. The book names the prerequisites correctly; the independent evidence says the majority of firms that fund the prerequisites still do not reach the top quartile. See research/04-consulting-firms/mckinsey-methodology-critique.md for the firm-level critique of McKinsey’s measurement approach.

Tension 2 — “Adoption and scaling” vs. workflow-first finding

Rewired places adoption/scaling as Capability 5/6, after technology and data. The independent corpus places workflow redesign as the binding constraint (see wiki/workflow-redesign.md):

McKinsey’s own State of AI 2025 (n=1,993): 55% of high performers fundamentally redesigned workflows when deploying AI vs. 18% of others.
BCG AI at Work 2025 (n=10,600): 70% of total AI value resides in people, organization, and process design.
Stanford Enterprise AI Playbook 2026: AI-primary workflows capture 71% median productivity gain vs. 30% for human-primary — same tool, different workflow architecture.

Rewired Ch. 5 (agentic workflow reimagination) and Ch. 30 (make adoption stick) do cover this — but the sequencing in the book’s six-capability structure can mislead readers into treating adoption as a Phase 3 concern. In practice, workflow redesign is a Phase 1 concern and must run parallel with data architecture investment, not after.

Tension 3 — “Build an AI platform” vs. PwC/Grant Thornton cost-to-value evidence

Rewired Ch. 21 and Ch. 22 advocate building an AI platform and an agentic engineering platform. For Fortune 500 with scale, this is plausibly economic. For the mid-market (the primary audience for State of AI briefings):

PwC AI Performance Study 2026: median mid-market AI build achieves 0.7–1.3x ROI over three years; median mid-market buy achieves 1.4–2.2x.
Grant Thornton 2026: 60%+ of mid-market AI platform builds are never fully decommissioned from vendor dependency — the “build” is net-new cost layered on top of existing license spend.
The corpus’s data-reset decision framework (see wiki/data-readiness.md) puts the Reset path — equivalent to Rewired’s full platform build — at 9–24 months and $450K–$2.3M even for mid-market single-workflow scope.

Synthesis: Rewired’s platform advocacy is correct for Fortune 500 / G2000 scale. For mid-market, the correct adaptation is (a) buy the platform, (b) invest in workflow redesign and data products on top, © reserve build-side investment for the one or two workflows that are competitively differentiating.

Recommended use in client engagements

Board briefings: Use Rewired’s “five transformation sins” (Introduction) as a diagnostic lens. Most Fortune 500 boards recognize three of the five on sight.
CEO / CIO alignment sessions: Ch. 3 (domains, not use cases) and Ch. 4 (business leaders lead) are the strongest single-chapter reads for a CEO who has delegated the transformation.
Data architecture investment cases: Ch. 25 (no data architecture, no AI advantage) plus the corpus evidence on data readiness (60% of AI projects abandoned through 2026 per Gartner) is the paired argument.
Agentic AI selection: Ch. 5’s four-levels framework and Exhibit 5.3 (best tool per task step) give a defensible, non-vendor-aligned way to say “no, this workflow is better automated with RPA + analytical AI than with agents.”
Caveats to surface: always cite the independent counter-evidence (NBER 2026, MIT CISR, METR) so clients don’t anchor on Rewired’s success cases as the expected outcome.

workflow-redesign — the binding constraint; Rewired Ch. 5 and Ch. 30 are supporting evidence
data-readiness — Rewired Ch. 25 (“no data architecture, no AI advantage”) is the data-backbone frame
ai-roi-evidence — Rewired adds E.ON Next as a Tier-2 named deployment; independent evidence continues to bracket the distribution
agentic-ai-governance — Rewired Ch. 34 (right to deploy) complements Bain’s Phase 1 framework
ai-center-of-excellence — Rewired Ch. 12 pods / Ch. 13 embedded engineering inform CoE design
ai-maturity-models — Rewired’s six-capability structure reads as a parallel to maturity models; cross-reference carefully
hitl-deployment-pattern — Rewired Ch. 5 (“humans remain essential, but their roles will evolve”) is the book’s HITL stance
ai-delivery-pods — Rewired Ch. 12–13 (pod model, embedded engineering); the delivery unit that operationalizes the operating model capability
mlops-ai-platform-engineering — Rewired Ch. 21 (AI platform) and Ch. 22 (agentic engineering platform); evaluation discipline from Ch. 5
data-products-reuse — Rewired Ch. 25 (“no data architecture, no AI advantage”), Ch. 26 (data products as reusable building blocks), Ch. 31 (“the best use case is the reuse case”)
ai-use-case-selection — Rewired Ch. 3 (domains over use cases), Ch. 5 (four-level agentic automation selection), Ch. 31 (reuse case thesis); Capability 1 deep dive
ai-talent-capability-building — Rewired Ch. 8 (business-leader tech fluency), Ch. 9 (IT talent surgery), Ch. 11 (workflow designer / agent supervisor / eval engineer); Capability 2 deep dive
ai-adoption-scaling — Rewired Ch. 30 (make adoption stick), Ch. 31 (design for scale and reuse), Ch. 32 (KPI discipline), Ch. 33 (midstream adjustments); Capability 5 deep dive