The Vertical AI Illusion: Why Generalists Will Win the Agent Economy
There’s a fundamental flaw in the vertical AI thesis: vertical AI is trained on subsets of enterprise data, but the context required for decision-making spans enterprise systems, and only horizontal platforms have access to that cross-functional data.
This isn’t about technology capabilities or domain expertise. It’s about data architecture and partnership economics. Horizontal platforms will win the orchestration layer while vertical AI succeeds only in regulated niches where compliance requirements create genuine moats. The market is dramatically overestimating the defensibility of vertical AI and underpricing the structural advantages of horizontal consolidation.
The Data Context Problem
Vertical AI companies build impressive domain-specific models. Harvey for legal reasoning, Hippocratic AI for clinical workflows, Duvo for retail operations. But these models face a structural constraint: they’re trained on domain-specific data while enterprise decisions require cross-functional context.
Consider how a CFO evaluates a contract. The legal AI can assess risk and compliance but that’s only one dimension. The CFO also needs financial exposure analysis, vendor relationship history, regulatory compliance status, and strategic alignment with ongoing initiatives. That context lives across Salesforce, NetSuite, SharePoint, compliance databases, and email archives. The legal AI sees one dimension; horizontal platforms orchestrate across all of them.
This creates a fundamental asymmetry: vertical AI trains deeply on narrow domains, but enterprise value creation occurs at the intersections of those domains.
The 2025 enterprise AI market reached $37 billion—with horizontal AI capturing $8.4 billion (5.3x YoY growth) versus vertical AI’s $3.5 billion (~3x growth). The gap isn’t narrowing; it’s accelerating. This isn’t random; it reflects the fundamental economics of data access.
Partnership Economics Create a Structural Moat
The partnership dynamics reveal why horizontal platforms compound advantage over time. Salesforce maintains 7,000+ ISV applications through 4,796 partner companies. Microsoft’s commercial ecosystem generates 95% of the company’s revenue and grows by 7,500 partners each month. Glean already supports 100+ native connectors despite being the youngest horizontal player.
Why does this matter? Because each integration creates a flywheel effect. Partners build platforms with maximum distribution; those platforms attract more partners, and customers benefit from comprehensive ecosystems, which in turn attract more partners. The ecosystem becomes the moat.
Vertical AI faces inverted economics. Instead of maintaining hundreds of integrations that unlock broad market access, vertical companies need deep integration with 3-10 core industry-specific systems. Each costs $50,000-$200,000 to develop, and annual maintenance accounts for 15-30% of the original build cost. Organisations underestimate integration costs by 40-60% on average, and 49% identify integration as their primary bottleneck to AI scaling.
The mathematics get worse: revenue-sharing agreements favour platforms. Salesforce ISV partners pay 15% of net revenue (dropping to 10% above $20M). For vertical AI, that same 15% applies to a dramatically smaller addressable market. A vertical AI company in legal tech might partner with 50 law firms; Salesforce partners access 157,000+ customers across every industry. Same integration cost, 3,000x difference in market access.
Enterprise Procurement Patterns Accelerate Consolidation
The strategic question facing boards: are enterprises expanding their vendor portfolios or consolidating them? The data is unambiguous. NPI Research’s 2025 survey shows 82% of enterprises actively pursuing supplier reduction, with 66% concentrating 80% of IT spend among 25 or fewer vendors. ADAPT’s CIO Edge research found that 68% of CIOs plan to consolidate vendors, targeting a 20% reduction in vendor count.
Why the consolidation imperative? Because every additional vendor creates exponential overhead. Security reviews for each vendor. Contract negotiations. Integration budgets. Ongoing vendor management. Support ticket routing. Compliance attestation. The administrative tax scales non-linearly with vendor count.
This structural reality creates a catch-22 for vertical AI: the more specialised and valuable the solution, the more it needs enterprise-wide integration, but integration is precisely what enterprises are trying to reduce. Andrew Ferguson of Databricks Ventures predicts “2026 will be the year CIOs push back on AI vendor sprawl.”
The platform bundling accelerates this dynamic. Microsoft bundled Sales, Service, and Finance Copilots at no additional cost in October 2025. Salesforce’s Agentforce 360 connects humans, agents, data, and metadata on a single platform. When horizontal platforms include comparable functionality, the procurement conversation shifts from “best-of-breed vs. integrated suite” to “why pay separately for narrow functionality our platform vendor already provides?”
Vertical AI must either be 10x better to justify separate procurement or operate in domains where horizontal platforms genuinely cannot compete.
Cross-Functional Orchestration Requirements Favour Horizontal Architecture
Enterprise workflows inherently span multiple domains, creating a structural need for horizontal orchestration that vertical AI cannot satisfy on its own. A typical sales workflow requires CRM data, email context, meeting notes, contract management, and financial approval across Salesforce, Outlook, Teams, DocuSign, and NetSuite. Vertical AI handles one component; horizontal platforms orchestrate the entire flow.
Menlo Ventures found 80% of enterprises starting with a single AI agent plan to orchestrate multiple agents within two years, yet fewer than 10% have successfully achieved multi-agent orchestration. This orchestration gap is the core value proposition of horizontal platforms. Gartner predicts that by 2028, “AI agent ecosystems will enable networks of specialised agents to dynamically collaborate across multiple applications and business functions.”
The data liquidity advantage compounds over time. Glean builds organisation-specific knowledge graphs that understand “relationships between people, content, and interactions” across dozens of enterprise systems simultaneously. Microsoft Copilot leverages Microsoft Graph to synthesise data from SharePoint, Teams, OneDrive, Exchange, and Dynamics in a single query. Salesforce’s Data 360 provides a “trusted, unified data layer that gives every agent context” across the entire CRM footprint.
Vertical AI operates in domain silos; horizontal platforms unlock cross-system insights that point solutions cannot achieve independently.
Where Vertical AI Maintains Genuine Defensibility
The counterargument deserves rigorous examination: regulated industries create genuine moats where vertical AI’s domain-specific precision justifies separate procurement and integration costs.
Harvey (legal, $8B valuation) serves 50 of the top AmLaw 100 firms precisely because legal work demands precision, horizontal tools cannot deliver. ChatGPT infamously fabricated six non-existent judicial decisions in the Schwartz case, a failure mode. Harvey’s domain-specific training prevents this through legal-specific guardrails and validation. Hippocratic AI (healthcare, $3.5B valuation) achieves 99.38% clinical accuracy with its Polaris 3.0 system, outperforming GPT-4 and Llama-3.1 by 74% on medical benchmarks through a 22-model constellation architecture specifically designed for clinical workflows.
The compliance architecture argument is legitimate. HIPAA certification, SOC 2 attestation, industry-specific audit trails, and regulatory guardrails must be architected into vertical AI from inception they cannot be retrofitted onto horizontal tools. Vertical AI startups also demonstrate superior GTM efficiency, with median sales and marketing spend of just 17% of ARR compared to roughly 50% higher for horizontal approaches.
Bessemer Venture Partners argues vertical AI can access labour budgets representing 13% of GDP versus just 1% for traditional software a 10x larger addressable market. Their portfolio data shows vertical AI companies growing approximately 400% year-over-year while achieving 80% of traditional SaaS contract values.
For boards evaluating vertical AI investments, the critical due diligence question is: Does regulatory complexity create a genuine moat, or is it a temporary advantage before horizontal platforms build equivalent compliance capabilities? Harvey and Hippocratic AI likely represent the former; most vertical AI represents the latter.
The Platform Orchestration Endgame
Despite vertical AI’s advantages in regulated niches, the strategic trajectory points toward horizontal platforms winning the orchestration layer and capturing the majority of enterprise AI value. The key insight: vertical AI agents will persist and potentially thrive, but they will increasingly operate through horizontal orchestration platforms rather than as standalone solutions.
Salesforce’s AgentExchange launched in March 2025 with 200+ partners specifically to enable specialised agents to plug into its Agentforce platform. Microsoft’s Agent 365 provides unified control for enterprise agents from Adobe, Databricks, Glean, NVIDIA, SAP, ServiceNow, and Workday, as well as open-source agents from Anthropic, LangChain, and OpenAI. Both support native MCP (Model Context Protocol) and A2A (Agent-to-Agent) protocols for interoperability.
The adoption metrics reflect this trajectory. Salesforce Agentforce reached $500M standalone ARR ($1.4B with Data 360) with 330% YoY growth—the “fastest-growing Salesforce product ever.” Microsoft Copilot achieved 70% adoption among Fortune 500 companies. Glean surpassed $100M ARR in under three years. Enterprise concerns about fragmentation, integration costs, and vendor management continue to intensify.
The endgame isn’t horizontal OR vertical, it’s horizontal orchestration WITH vertical specialisation running through platform rails.
Strategic Implications for Boards and Investors
Boards evaluating AI strategies face a capital-allocation decision: bet on domain-specific point solutions or on horizontal platforms that orchestrate across the entire technology stack? The evidence suggests horizontal platforms for core infrastructure, with vertical AI deployed selectively where regulatory moats create genuine defensibility.
The optimal architecture emerging: horizontal orchestration layers coordinating specialised vertical agents. This is precisely what Salesforce and Microsoft are building, not monolithic platforms that do everything, but orchestration layers that coordinate specialised capabilities while maintaining unified data access and governance.
For vertical AI companies, the strategic path depends on moat durability. Companies with exceptional domain advantages in regulated industries (legal, healthcare, financial services) may sustain independence. Others will likely integrate into horizontal platform ecosystems either through partnerships, OEM arrangements, or acquisition.
For investors, the thesis bifurcates: Bessemer and Index Ventures’ bullish vertical AI positioning remains valid for regulated niches, but the platform layer consolidates around horizontal incumbents. The arbitrage opportunity exists precisely because vertical AI companies will eventually partner with or be acquired by horizontal platforms creating attractive M&A targets but challenging standalone public market trajectories.
The fundamental question: in an agent economy requiring cross-functional orchestration and enterprise-wide data context, will you optimise for domain-specific point solutions or horizontal platforms that orchestrate your entire technology ecosystem? The partnership economics, procurement patterns, and data architecture requirements increasingly favor horizontal consolidation with vertical AI persisting only where regulatory complexity creates genuine, defensible moats.
What’s your perspective on the horizontal vs. vertical positioning debate?
Originally published on LinkedIn.