--- title: "Vertical AI" canonical: "https://anchor.is/on-deck-vvSq/issue/15/vertical-ai" publication: "On Deck" publisher: "Ground Truth" language: "en-US" guide: "https://anchor.is/on-deck-vvSq/llms.txt" permalink: "https://anchor.is/on-deck-vvSq/issue/15/vertical-ai" issue: 15 section: "vertical-ai" description: "How AI is reshaping industries, company by company" authors: ["Reva Sandoval"] published: 2026-10-12 updated: 2026-10-11 articles: 16 --- # On Deck > Ground Truth follows AI as it's actually built, funded, and shipped: consumer assistants, vertical products from legal and healthcare to agriculture, and the capital moving behind them. It's written for practitioners who've shipped agentic systems and want more than press releases. It reads interaction models, architectures, trust patterns, and funding logic to judge which companies are serious, and where it's worth building next. [Read on the website]() ## Issue 15 — Vertical AI How AI is reshaping industries, company by company ### Vertical Deep Dive - [Who Controls Legal AI: Workflow, Model, or Platform?]() — By [Reva Sandoval]() · Published 2026-10-11 · Format: Vertical Competitive Read Clio's acquisition of Learned Hand in September 2026 puts the company on both sides of the courtroom: tools for the lawyers filing, and tools for the judges reviewing what they filed. It extends an eight-year acquisition sequence worth reading against two companies running different theories of control in legal AI. Harvey owns the model layer and authored the benchmark other models are scored against. Google is packaging legal skills inside its enterprise platform. The comparison runs across workflow position, data authority, buyer, and evaluation ownership — and names the position hardest to replicate. ### Company Map - [Insurance AI's buyer split is the only map you need]() — By [Reva Sandoval]() · Published 2026-10-11 Insurance AI sorts by buyer. The product either sells to brokers or to carriers, and the two sides share almost nothing: different integration surfaces, different regulatory exposure, different capital dynamics. Learn this distinction and the vertical makes sense. Ignore it and every company blurs together. ### Vertical Scan Eight verticals, six weeks of signal. The cross-cutting pattern: capital is funding structurally incompatible bets within the same vertical. Platform versus proprietary data loop in finance. Clinician distribution versus autonomous clinical agency in healthcare. Three different control models in legal, all placed within three weeks. Nobody has converged on what the winning AI product architecture looks like in any of these industries. Two transactions were large enough to reshape competitive structure: Schneider's $22.6B PTC acquisition in industrial, and Harvey's $15.5B valuation in legal. One constraint mattered more than any product launch: Amazon blocking Meta's Muse established that merchants, not users, gate agentic commerce. The remaining verticals are staking positions in layers that have not consolidated yet. - **Three Incompatible Control Bets in Three Weeks** — [Source]() — By [Reva Sandoval]() — Published 2026-10-11 Tag: Legal AI Clio acquired Learned Hand for court-record ownership. Harvey closed $550M at $15.5B to own model post-training. Google launched Gemini Enterprise for Legal as a horizontal platform play. Each bet assumes a different layer is the one that matters. Deep Dive maps the positions. - **Private and Federal Capital Target Different Products** — [Source]() — By [Reva Sandoval]() — Published 2026-10-11 Tag: Healthcare AI OpenEvidence raised $250M at $15B on clinician distribution: over a million licensed U.S. users, embedded in Epic workflows. ARPA-H committed $62.7M to ADVOCATE, the first serious push toward an FDA-authorized, patient-facing cardiovascular agent. The private money is buying adoption. The federal money is buying regulatory precedent. Watch which one creates the durable moat. - **Golub's F2 Deal Shows the Data-Loop Advantage** — [Source]() — By [Reva Sandoval]() — Published 2026-10-11 Tag: Finance AI Golub Capital invested $5M in F2 and will deploy F2 agents against Golub's proprietary data for screening, underwriting, and portfolio monitoring. Google launched Gemini Enterprise for Financial Services with 50+ reusable skills and licensed-data connectors. The difference: F2's arrangement creates a feedback loop where agent performance improves on institutional data Google cannot access. No independent performance evidence is public yet for either. - **Schneider's $22.6B PTC Acquisition Carries Product Debt** — [Source]() — By [Reva Sandoval]() — Published 2026-10-11 Tag: Field & Industrial Schneider Electric agreed to acquire PTC for $22.6B, combining CAD, PLM, field service, and energy management. The complication: PTC runs two CAD systems and two PLM systems connected by interoperability layers, not a converged product model. Schneider is buying distribution and portfolio breadth, but also inheriting the integration problem PTC never resolved. Any AI layer built on top will reflect that fragmentation. - **Three Mid-Stack Moves, No Risk Authority Claimed** — [Source]() — By [Reva Sandoval]() — Published 2026-10-11 Tag: Insurance AI Outmarket raised $34.5M for broker paperwork automation. Adlib acquired Paperbox for agentic document intake. Sapiens added an agent layer over its core policy-administration products. All three sit in administrative and intake positions. None puts the vendor in the risk-bearing seat. Company Map builds the full landscape. - **Amazon Blocked Muse and Revealed the Real Gate** — [Source]() — By [Reva Sandoval]() — Published 2026-10-11 Tag: DTC/E-commerce AI Amazon blocked Meta's Muse from purchasing on its platform within two weeks of launch. The structural lesson: user authorization does not grant transaction permission. The merchant controls that gate. Visa's CEO confirmed the demand-side version of the same constraint: consumers use AI to browse but only 23% trust it with payments. Companies building merchant-side agentic tools hold the safer position right now. - **Deere Launches JD as FBN Builds Competing Context** — [Source]() — By [Reva Sandoval]() — Published 2026-10-11 Tag: Agriculture AI John Deere released JD, an AI assistant inside Operations Center, in limited early access for U.S. farmers. The quieter move: Farmers Business Network partnered with Google AI Futures Fund to build an AI Context Engine outside Deere's ecosystem. The vertical's structure depends on whether Deere's telemetry advantage compounds faster than farmer-owned networks can aggregate equivalent context independently. - **Accounting AI Splits Into Three Non-Overlapping Layers** — [Source]() — By [Reva Sandoval]() — Published 2026-10-11 Tag: Professional Services Accrual launched Arc as a firm-wide agent platform. Crunchafi acquired BUCS Analytics for 150+ system integrations and client-data access. Numeral raised $100M Series C for bounded sales-tax compliance. Three companies occupying three distinct positions with no published accuracy evidence for any of them. The Numeral round is the signal: $100M says investors believe high-frequency compliance is defensible at scale. ### Further Reading - **Harvey LAB Initial Results: Frontier Models Complete Less Than 10% of Legal Tasks End-to-End** — [Source]() — Published 2026-10-11 Harvey's own benchmark data on where frontier models fail at legal work. The cost-per-task numbers are revealing. - **Harvey Tenet: Post-Training an Open-Weight Model for Legal Work on 150 B300 GPUs** — [Source]() — Published 2026-10-11 How Harvey fine-tuned Kimi K3 for legal reasoning and why they chose open-weight over proprietary models. - **Google Gemini Enterprise for Legal: Full Platform Architecture and Connector Ecosystem** — [Source]() — Published 2026-10-11 Google's legal AI platform spec — skills, MCP connectors, governance controls, and launch customers including Harvey as partner. - **Clio Acquires Learned Hand, Expanding Into Judiciary AI** — [Source]() — Published 2026-10-11 Clio's acquisition rationale and Learned Hand's existing court deployments. Watch the billion-document data angle. - **Google's GC on AI Limits: 2,000+ Hallucination Cases and the Billing Model Problem** — [Source]() — Published 2026-10-11 Google's own general counsel on where legal AI breaks — hallucination rates, confidentiality risk, junior lawyer displacement. - **ARPA-H ADVOCATE: The First FDA-Authorized Clinical AI Agent Program** — [Source]() — Published 2026-10-11 Federal program funding autonomous cardiovascular AI agents with Tempus and Kaiser. First FDA-authorized clinical agent pathway. ## Other sections in Issue 15 - [Signal Watch]() - [Horizontal AI]() - [AI Capital]() ## More - [Publication guide]() - [Issue archive]()