How to read this
Rows = paper sections. Columns = source firms / authors. Cells = the specific piece I'll cite. Bold indicates a load-bearing citation (the section's argument depends on it). Italics indicates a flagged gap — no strong secondary source found, original analysis carries the load. Dates are publication dates, not access dates.
Citation discipline: every quantitative claim gets a primary or near-primary source (10-K, S-1, vendor press release, industry analyst report, dated newsletter). Every framework claim gets the canonical author. AI-era pressure tests prefer 2023–2026 material; canonical moat theory uses originals (Porter 1979, Helmer 2016, Greenwald 2005).
Theory chapters (Part I)
| Section | Helmer / canon | NfX / a16z / Stratechery | HBR / academic | AI-era empirical |
|---|---|---|---|---|
| Ch. 1 — Scale Economies | Helmer, 7 Powers, ch. 1 (2016) | Stratechery, "Aggregation Theory" (2015) | Porter, "Competitive Strategy" (1980) | Epoch AI, "Trends in machine learning compute" (2024); SemiAnalysis on inference cost curves (2024–2025) |
| Ch. 2 — Network Economies | Helmer, ch. 2 | NfX, "Network Effects Manual" (16 effects); NfX "70% of value in tech" | Eisenmann/Parker/Van Alstyne, "Strategies for Two-Sided Markets," HBR (2006) | Sacra company breakdowns on multi-tenant network density |
| Ch. 3 — Counter-Positioning | Helmer, ch. 3 (the most original chapter; load-bearing) | Stratechery on Microsoft's seat-based pricing risk | Christensen, Innovator's Dilemma (1997) — adjacent but not identical | Gap — original analysis on AI counter-positioning against per-seat SaaS. Use Sierra, Decagon, Crescendo as case evidence. |
| Ch. 4 — Switching Costs | Helmer, ch. 4; Greenwald, ch. 5 (customer captivity) | NfX "embedding" defensibility | Klemperer, "Markets with Consumer Switching Costs," QJE (1987) — academic anchor | Gartner ERP migration cost benchmarks; vendor case studies on SAP/Oracle migrations |
| Ch. 5 — Branding | Helmer, ch. 5 | NfX, "What's Your Defensible Magic" | Keller, Strategic Brand Management (canonical) | Edelman Trust Barometer 2024–2026 on AI-era trust; gap — no clean source on "branded calm" / trust as scarce in generative volume; original argument. |
| Ch. 6 — Cornered Resource | Helmer, ch. 6 | — | Barney, "Firm Resources and Sustained Competitive Advantage" (1991) — RBV anchor | SemiAnalysis / The Information on GPU allocation contracts; coverage of OpenAI / Anthropic data licensing deals |
| Ch. 7 — Process Power | Helmer, ch. 7 | Stratechery on Toyota / TSMC process | Womack, Jones, Roos, The Machine That Changed the World (1990) | Gap — limited published material on AI's effect on Process Power. Original argument carries; supporting case from McKinsey "State of AI" 2024. |
| Ch. 8 — Data Flywheel | — | a16z, "The Empty Promise of Data Moats" (Casado/Bornstein, 2020); a16z "Trading Margin for Moat" (2024); NfX "AI Defensibility" (2023+) | — | Sequoia, "AI 50: Agents Move Beyond Chat" (2025); Tesla / Waymo coverage from The Information; Ferguson Analytics, "Data Moats in the AI Era" (2024) |
| Ch. 9 — Agentic Workflow Lock-in | — | NfX on personal-utility network effects; a16z on services-led growth and agent moats (2024–2025) | — | Sacra company memos on Cursor, Sierra, Glean, Decagon (2024–2026); flag for Priya: technical durability of agent memory under MCP standardization. |
Master matchup matrix (Part II)
The matrix itself is original synthesis. Supporting evidence per claim drawn from Part I and Part III sources. The three "loud claims" defended in prose are anchored to the paper's 4 load-bearing asterisks in Part III.
| Claim | Primary support | Counter-argument source | Anchored asterisk |
|---|---|---|---|
| Data Flywheel beats Distribution once spinning | a16z "The Empty Promise of Data Moats" (the nuanced version); Tesla/Waymo coverage; Glean's 12–18-month maturation curve | a16z's own caveat that most "data moats" never reach escape velocity | Glean vs Microsoft Copilot at H2 |
| Switching Costs beat AI-Tech Differentiation in regulated/embedded domains | Gartner ERP migration data; Helmer ch. 4; Bloomberg terminal case | Stratechery on Microsoft Copilot's distribution-based incursion; agentic-execution-layer case (Decagon/Crescendo) | NetSuite vs AI-native ERP greenfield at H2 |
| Counter-Positioning strengthens with incumbent AI investment | Helmer ch. 3; Sierra/Decagon counter-positioning vs. legacy CX SaaS; Speckle vs Autodesk file-format trap | Microsoft's Copilot bundling as evidence that incumbents can sometimes retrofit | Speckle vs Autodesk at H2; Suffolk vs owner-led IPD at H2 |
Industry war games (Part III)
Chapter 10 — AEC general contractors
| Source | Specific piece | Use |
|---|---|---|
| Construction Dive | "Rubber hits the road on AI startups: Suffolk tech lead" (2024); Suffolk Technologies fund coverage | Suffolk's bet pattern |
| ENR / Engineering News-Record | "Turner Construction Launches Turner Ventures" (2024) | Turner CVC strategy |
| Data Center Dynamics | "Turner Construction doubles data center revenue in 2025" | What scale-economy actually buys a GC in the AI buildout |
| BD+C / Building Design + Construction | "AI for construction" overview (2024) | Industry-level adoption baseline |
| McKinsey Global Institute | "Reinventing construction" (2017, still the canonical productivity baseline) + 2024 follow-ups | Industry-productivity context |
| BCG / Bain | Gap — limited recent construction-tech-specific reports from BCG/Bain. Use Roland Berger or Deloitte construction-tech reports as substitute. | Cross-check |
Chapter 11 — AEC software (Autodesk + 4 attack archetypes)
| Source | Specific piece | Use (which archetype) |
|---|---|---|
| AEC Magazine (Martyn Day) | "Contract killers: how EULAs are shifting power" on Autodesk; "Snaptrude on AI" (2025); "Snaptrude AI: conceptual design and beyond" (2025); ongoing coverage of generative-design tooling | Load-bearing for entire chapter. Autodesk contrast probe + Archetype B (file-format). |
| Autodesk 10-K + earnings calls | FY24, FY25, FY26 filings; transcripts on subscription / transaction-model shift; segment commentary on AI roadmap | Autodesk seat economics, revenue concentration, stated AI strategy. Load-bearing for the contrast probe. |
| BeyondSPX / equity research | "Autodesk's Direct Revolution" on transaction model; sell-side notes on cannibalization risk | Autodesk defensive moves — the contrast probe argument. |
| Klover.ai / industry analyst notes | "Autodesk's AI Strategy" overviews; Roland Berger / Deloitte AEC-tech analyses | Autodesk's stated AI roadmap and adoption baseline. |
| Speckle Systems | Open-source repo activity; "Changing the Way We Work in AEC with Snaptrude and Speckle"; case studies | Archetype B · file-format / data-translation. Speckle commercial layer + open-source credibility. |
| Higharc press / coverage | Funding rounds, customer announcements (homebuilder partnerships); ENR / Builder magazine coverage | Archetype A · contract / commercial-relationship disruption. |
| Trunk Tools, Document Crunch press / Sacra (where available) | Customer counts, ARR, founder podcasts (Latent Space, Practical AI on AEC); Construction Dive coverage | Archetype C · document / spec / risk AI agent. |
| Buildots, OpenSpace press / case studies | Skanska / Suffolk / Turner deployment case studies; Construction Dive + ENR coverage; gap — no third-party benchmark on flywheel-data accumulation rate per customer (flagged below) | Archetype D · field-capture data flywheel. |
| Nvidia OpenUSD / hyperscaler standards push | Nvidia GTC announcements 2024–2026; OpenUSD adoption coverage | Counter-attack vector against Speckle's open-data positioning (asterisk-adjacent, not the asterisk itself). |
Chapter 12 — Accounting
| Source | Specific piece | Use |
|---|---|---|
| Intuit Investor Relations | "Accountants Embrace AI: 2025 Intuit QuickBooks Survey"; FY25 annual report | Intuit's 80% SMB share, GenOS, fine-tuned financial LLMs |
| Intuit Enterprise Suite blog | "AI-Native ERP for Mid-Market Growth" (2024) | Intuit's mid-market push directly at NetSuite |
| Oracle / NetSuite | NetSuite 2024.1 / 2024.2 release notes; SuiteCentric "NetSuite AI Ultimate Guide 2025" | NetSuite's 200+ AI features claim |
| CFO Dive | "Intuit rolls out AI QuickBooks enhancements" (2025) | Agent rollout pace |
| Puzzle.io / VentureBeat | "Brex turns accounting into a one-click setup with Puzzle integration" | AI-native attacker stack |
| Sacra | Pilot, Puzzle, Brex, Ramp company memos | Revenue/customer triangulation |
| Gartner Magic Quadrant — Cloud ERP | 2024, 2025 editions | NetSuite vs. competitors positioning |
Chapter 13 — AI-native horizontal (4 attack archetypes)
| Source | Specific piece | Use (which archetype) |
|---|---|---|
| Sacra | Glean, Hebbia, Cursor, Harvey, Sierra, Decagon, Crescendo, Clay company memos (2024–2026) | Load-bearing for revenue/growth claims across all 4 archetypes. Sacra is the most rigorous tracker of private AI-native ARR. |
| Glean press / blog; Futurum Group | Series F announcement ($7.2B, $200M ARR, June 2025); "Glean Doubles ARR to $200M. Can Its Knowledge Graph Beat Copilot?" | Archetype 1 (horizontal knowledge / retrieval). Microsoft/Copilot threat framing — the load-bearing matchup. |
| The Information | Coverage of Cursor ($2B ARR), Harvey ($8B val), Sierra ($10B), Decagon rounds (2024–2026) | Mid-market deal context across all 4 archetypes. |
| Sequoia Capital | "AI 50: AI Agents Move Beyond Chat" (2025); ongoing market-map updates | Cohort framing; Archetypes 2 + 4 (vertical co-pilots; agentic execution). |
| Upstarts Media | "Decagon, Sierra AI And The Race To Build Customer Support Agents" | Sierra/Decagon head-to-head — Archetype 2 vs Archetype 4 boundary case. |
| CNBC / Crunchbase News | Glean Series F (June 2025); Cursor / Harvey / Clay round coverage | Pricing-the-round context. |
| Microsoft 10-K + earnings; Stratechery | M365 Copilot adoption stats; "AI Bundles" essays; commentary on Microsoft enterprise AI distribution | The distribution side of the central asterisk: Glean vs Microsoft Copilot at H2. |
| Latent Space podcast / The Generalist | Founder interviews — Cursor, Harvey, Sierra, Decagon, Crescendo (2024–2026) | Triangulating attack-vector claims (UX-rebuild vs domain-data vs agentic execution). |
Cross-cutting AI-era frame
| Source | Specific piece | Use across paper |
|---|---|---|
| McKinsey | "The State of AI in 2024" / 2025 follow-up | Adoption baselines, function-level penetration |
| BCG | "AI Adoption Index" 2024–2025 | Maturity-curve framing |
| Bain | "Technology Report" annual (2024, 2025) | Vendor-landscape framing |
| Gartner | Hype Cycle for Generative AI 2024, 2025; Magic Quadrants for ERP, CRM, Enterprise Search | Vendor positioning, expectations frame |
| HBR | "Competing in the Age of AI" — Iansiti & Lakhani (2020); follow-up pieces 2023–2025 | Theory anchor for "AI factory" concept; cite carefully — predates GenAI specifics |
| Stratechery (Ben Thompson) | Aggregation Theory; "AI Bundles" (2024); pieces on Microsoft / Apple / Adobe AI strategy | Distribution-side reasoning |
| The Information / The Generalist / Latent Space podcast | Ongoing AI-company coverage | Triangulating private-company claims |
Flagged gaps
Five sections where I couldn't find a strong secondary source and original analysis will carry the argument. Calling these out so Dania can pressure-test rather than discover late.
- Counter-Positioning in the AI era. Helmer's chapter is excellent but pre-AI. No clean secondary essay specifically on "AI counter-positioning vs. seat-based SaaS." Argument is original; case evidence is Sierra / Decagon / Crescendo.
- Trust-as-moat under generative-AI volume. Edelman Trust Barometer is too generic. The "branded calm" argument (in a world of AI-generated sludge, trusted brands become more valuable, not less) is original.
- Process Power under AI pressure. No published essay attacks this directly. Original argument: AI is the most direct attack on Process Power because it lowers the cost of replicating tacit organizational know-how — but this only works in domains where the know-how is decomposable into model-trainable patterns. Hand-off flag for Priya.
- AEC GC moats. The construction-tech press covers tech adoption, not the underlying moat structure of GCs. The chapter's central question — "do Suffolk/Turner/DPR have moats at all, or just bonding capacity + relationships?" — is original analysis. Will use S&P / Moody's bonding capacity data + ENR top-400 rankings as supporting evidence.
- Agentic Lock-in durability. Too new for a strong literature. Pass-2 hand-off to Priya for the technical durability question. Commercial side will lean on case evidence (Cursor's project-context stickiness, Glean's 12–18-month knowledge-graph maturation claim).
- AEC field-capture flywheel — data-accumulation rate. Buildots and OpenSpace publish customer counts and a few case studies, but no credible third-party benchmark exists on (a) how much proprietary data each flywheel actually accumulates per customer per year, and (b) whether that data has decreasing or increasing marginal value past year 1. The Archetype D flywheel thesis depends on this. Pass 2 treats this archetype as "thesis-grade, not benchmarked" with an explicit "I'd update toward [X] if I saw [Y]" disclosure rather than asserting escape velocity it can't yet prove.
- Autodesk generative-design defensibility. The Autodesk contrast probe argument is materially shaped by whether generative design clears "production-ready BIM" by H2 = 3 years, or stalls at concept design. No clean public source predicts this. Hand-off flag for Priya before Pass 2. If she says "stalls at concept," the Autodesk argument softens; if "production-ready by H2," the argument hardens.
What I will not cite
- LinkedIn thought-leadership posts (signal too noisy)
- Generic "top 10 AI startups" listicles (no analytical value)
- Vendor-sponsored research without the sponsor disclosed (e.g., AI-vendor-funded "AI ROI" reports)
- Anything I can't trace to a primary source within two clicks