Intelligence applied to specific verticals.
“The $11 trillion U.S. labor spend dwarfs the $450 billion enterprise software market. Applied AI is the first technology positioned to capture a meaningful share of that labor budget.”
Foundation models are commoditizing. The wealth creation in this cycle is shifting decisively from the model layer to the application layer. Bessemer's analysis is clear: vertical AI targets high-cost, language-based tasks that were unreachable by prior generations of software. The total addressable market is not the existing software budget. It is the labor budget.
Data flywheel effects create moats that general-purpose models cannot replicate. Harvey's legal knowledge graph improves with every case it processes. Glean's enterprise search becomes more valuable with every document it indexes. These are not thin wrappers around foundation models. They are domain-specific intelligence systems that accumulate proprietary advantage with every interaction.
We invest in applied AI companies that have identified a specific wedge into a high-value vertical, are accumulating proprietary data through their workflow, and are building systems of record rather than copilots. The companies that become essential to their users' workflows will capture the largest share of the labor-to-software transition.
01
Harvey's valuation trajectory -- $3 billion to $8 billion to $11 billion in under twelve months -- is unprecedented in enterprise software. Glean reached $100 million ARR in under three years at a $7.2 billion valuation. EvenUp crossed $2 billion. These are not outliers. They are the new benchmark for what category-defining AI companies look like.
02
The agentic AI market reached $7.6 billion and is projected to hit $52 billion by 2030. Anthropic is deploying enterprise agents through partnerships with PwC. The shift from copilot to autonomous agent represents a step change in the value AI can deliver per dollar of software spend.
03
Carta's data shows AI-native companies raising Series A rounds at $22 million median, versus $15 million for traditional SaaS. The market is pricing in the structural advantage of AI-native architectures. Horizontal tools that bolt AI onto existing workflows are losing to vertical tools that rebuild workflows around AI.
04
Writer reached $1.9 billion valuation with 194% ARR growth. Cohere hit $240 million ARR and is anticipated to IPO. The companies that have accumulated domain-specific training data and fine-tuned models on it have a compounding advantage that cannot be replicated by throwing more compute at a general-purpose model.
Industry-specific software where AI is native to the architecture, not an add-on. Veeva in life sciences, ServiceTitan in trades, Procore in construction, Clio Duo in legal. These companies own the workflow and the data, making their AI capabilities improve with every customer interaction. The moat is the domain-specific intelligence that accumulates over time.
AI systems that automate specific high-value workflows. FurtherAI raised $25 million Series A for insurance automation. SmarterDx applies AI to clinical documentation. EvenUp automates personal injury demand letters. Abridge transcribes and summarizes clinical conversations. Each targets a workflow where manual labor costs thousands of dollars per case.
Tools that augment professional judgment rather than replace it. Glean, at $7.2 billion, is the enterprise search layer. Writer, at $1.9 billion, generates compliant content at enterprise scale. Sierra, valued at $10 billion with $350 million raised, builds AI customer experience agents. Cohere provides the model layer for enterprise AI applications.
Autonomous systems that complete multi-step tasks. Hippocratic AI, valued at $3.5 billion after a $126 million Series C, builds healthcare agents. Sierra, Harvey, and EvenUp are deploying agents that can manage entire case workflows. The shift from copilot to agent expands the value capture from productivity gains to labor replacement.
“We invest in outcomes, not copilots.”
Domain-specific data that makes models better than general-purpose alternatives. Every interaction should make the system more valuable. If removing the AI breaks the data flywheel, you have a real moat.
Cases settled, revenue recovered, time-to-decision reduced. We look for companies that can quantify their impact in units their customers already track, not in abstract productivity metrics.
Removing the AI breaks the workflow. System of record, not copilot. When the tool becomes the surface through which work happens rather than a suggestion layer, switching costs become structural.
Founders who have lived the problem they are solving, or deep domain partnerships that provide the equivalent insight. The best applied AI companies are built by people who understand the workflow before they write the first line of code.
Security certifications, data residency controls, and audit trails. In healthcare, legal, and financial services, trust infrastructure is a prerequisite, not a roadmap item.
Early signs of expansion across workflows within existing customers. The best applied AI companies start with a single use case and become the operating system for an entire department.
Building applied AI
for a specific vertical?