Rise of Vertical AI Agents: Strategic and Investment Implications (1/2)

June 2025 

The Rise of Agentic AI: A Paradigm Shift in Enterprise Software

Having spent the better part of the last decade advising vertical SaaS firms on capital raising and M&A, one becomes attuned to the slow churn of enterprise technology. Trends emerge, mature, and consolidate. Occasionally, a technology disrupts the cost structure or sales cycle. Rarely does it rewrite the architecture of value itself. Yet that, increasingly, is what artificial intelligence, and more precisely, agentic AI, appears poised to do.

These agents are not merely another turn of the automation screw. They represent a categorical shift: from software-as-a-tool to software-as-a-doer. Where traditional SaaS enhanced workflows, AI agents autonomously execute them. Reasoning, decision-making, and iterative learning are no longer confined to human operators; they are becoming attributes of software. This is not the evolution of SaaS. It is its potential obsolescence.

What distinguishes this wave from prior technological shifts is its scope. Cloud computing unbundled IT departments and decentralised infrastructure, but it largely left services untouched. AI is disrupting both software and services at once. That is why its total addressable market is plausibly an order of magnitude larger than cloud’s ever was.

Moreover, the preconditions for scale are already in place. The internet is saturated with billions of connected users. Distribution is instant and viral, piggybacking on platforms like Reddit and X. Compute capacity, talent pools, and high-quality data are no longer bottlenecks. The post-ChatGPT world has compressed the adoption cycle: awareness, demand, and access have been fully activated. What remains is execution.

As in the SaaS era, the spoils will accrue not to the inventors of infrastructure, but to those who build applications atop it. While billions are being sunk into foundation models, it is in the messy, high-friction trenches of vertical markets, healthcare, logistics, law, where the next wave of enterprise champions is likely to emerge. The winning formula is becoming clear: solve complex problems with specificity; integrate human oversight to build trust; and deliver outcomes, not just cleverness.

For all the buzz, the buyers remain cautious. AI that fails to deliver is not new, it is merely expensive. In sectors where regulatory scrutiny is tight and workflows are mission-critical, the bar for adoption is high. Vendors that succeed will not do so by dazzling with generative prowess, but by demonstrating measurable returns on investment.

Perhaps most important, the agentic era demands a cognitive shift. Traditional software was deterministic: input in, output out. AI is probabilistic, fluid, and prone to hallucination. It forces managers to abandon certainty for confidence intervals, and engineers to design not just for success but for variability. Trust, in this new world, is earned not through control, but through coherence and repeatability over time.

SaaS reshaped enterprise computing. Agentic AI may reshape the enterprise itself.

Defining Vertical AI Agents: A New Software Paradigm

Over the past decade, enterprise software has evolved through major innovations, cloud computing, mobile-first design, and the widespread adoption of SaaS platforms. However, a more profound transformation is now underway, driven by the rise of agentic AI. At the forefront of this shift are vertical AI agents: specialized, autonomous systems designed for deep integration within specific industries or functional domains. These agents represent more than an incremental improvement over SaaS, they signal a foundational change in how software is conceived, deployed, and utilized across enterprises.

The emergence of vertical AI agents is best understood through a three-tier framework, mapping the progression from basic generative models to autonomous, context-aware actors.

At the first level are Large Language Models (LLMs), systems like ChatGPT, Claude, or Gemini. These are capable of generating human-like responses based on vast amounts of training data. Yet they are fundamentally reactive; they cannot access private or real-time data unless explicitly integrated, and they are incapable of initiating action. LLMs require human prompting for every task and thus function more as intelligent respondents than proactive actors.

The second level introduces AI Workflows. These are structured sequences in which LLMs follow predefined instructions to accomplish specific tasks, such as retrieving information via APIs or querying databases. While more functional than LLMs alone, these workflows remain dependent on manual configuration. Their logic and structure are designed by humans in advance, which limits adaptability and creative problem-solving. Tools like Make.com or techniques like Retrieval-Augmented Generation (RAG) fall into this category, useful, but still constrained.

The third and most advanced tier is AI Agents. These systems go beyond reactive intelligence and are capable of reasoning, taking autonomous action, and learning iteratively. Built atop foundational LLMs, AI agents can define goals, sequence actions, access external tools, evaluate outcomes, and refine their strategies, often without any human intervention. For example, an agent today can autonomously search and summarize news, draft a LinkedIn post, adjust tone and formatting for optimal engagement, and publish, all as part of a single autonomous workflow. With advancements in vision models, some agents are even capable of identifying and labeling visual content (like detecting a skier in a video), demonstrating multi-modal intelligence in production-grade tasks.

Vertical AI agents represent a distinct evolution in this hierarchy. Unlike horizontal agents, which offer general-purpose functionality across use cases, vertical agents are finely tuned to operate within the specific context of industries like law, medicine, tax, software development, or cybersecurity. They understand and embed the language, workflows, and regulatory constraints of their respective fields. This deep contextual intelligence allows them to perform entire workflows typically reserved for highly trained human professionals.

Several early examples underscore this shift: Harvey is transforming legal research and drafting; Abridge automates medical notetaking; TaxGPT is streamlining tax advisory; Devin and Traversal are pushing the frontier in software development automation; and Expo applies agentic logic to cybersecurity monitoring and remediation. These are not mere tools, they are autonomous contributors, executing tasks that once required expert human input.

The importance of vertical AI agents lies in their ability to deliver superior performance in domains where traditional AI has struggled. In regulated, high-stakes industries like healthcare, legal services, or financial compliance, domain knowledge and contextual precision are essential. Vertical agents outperform generalist models by being intimately familiar with sector-specific terminology, logic, and constraints, producing more accurate, relevant, and compliant output.

From a commercial standpoint, these agents offer a faster path to adoption and clearer returns on investment. By targeting well-defined problems with concrete metrics, such as documentation time, error rates, or turnaround speed, vertical agents deliver immediate operational benefits. Their deployment often results in faster proof-of-concept cycles, stronger customer retention, and lower cost of acquisition, especially in sectors hungry for automation.

Moreover, vertical agents are less exposed to the threat of commoditization. While horizontal AI tools face intense competition from Big Tech ecosystems (Microsoft Copilot, AWS Bedrock, OpenAI’s APIs), vertical agents carve out defensible moats. Their value derives not from access to generalized LLMs, but from integration with proprietary data, domain-specific workflows, and niche user requirements, elements that large platform players are less likely to replicate at depth.

Though horizontal agents benefit from broader applicability and larger total addressable markets, they face several limitations: slower adoption due to less defined use cases, higher business model risk due to commoditization, and lower defensibility. Vertical agents, while targeting narrower segments, enjoy faster adoption, stronger product-market fit, and higher user retention. Their deep integration into industry-specific processes makes them harder to displace and more aligned with enterprise buyers seeking tangible ROI.

Developers pursuing vertical agent models tend to adopt one of two main strategies, or a hybrid of both.

The first is agent orchestration and evaluation frameworks. Here, multiple modular agents are trained to collaborate by delegating tasks, monitoring performance, and dynamically reassigning responsibilities. This model supports composable and scalable systems, particularly important for complex enterprise applications where workflows span multiple interdependent roles.

The second strategy involves end-to-end fine-tuning. This approach focuses on training LLMs using proprietary, domain-specific data and workflows to create “out-of-the-box” agents that deeply understand the target environment. These fine-tuned agents can recognize patterns, interpret user needs, and execute complex tasks with little to no additional prompting, often outperforming generic models in both accuracy and utility.

Looking Ahead: From Agent Swarms to Agent Economies

As agentic technology matures, the future will not be defined by isolated agents, but by agent swarms, cooperative networks of AI entities capable of coordinating work, sharing data, and maintaining persistent identities across interactions. These swarms will resemble decentralized digital organizations, executing collaborative tasks and self-optimizing for performance.

Eventually, this may give rise to agent economies, where autonomous systems negotiate, transact, and prioritize tasks across organizational boundaries. In this world, agents will represent users, departments, or even entire companies, interacting through shared protocols and trust frameworks.

Enabling such a future will require advances in several areas. Persistent memory is essential so that agents can build long-term context and continuity across tasks. Standardized communication protocols, such as the emerging Machine Communication Protocol (MCP), will allow inter-agent interoperability. And most critically, robust trust and security layers will be needed to verify agent identity, ensure ethical behavior, and enable traceability in high-stakes decision-making environments.

In sum, vertical AI agents are not just the next generation of enterprise software, they are a foundational shift toward software that acts, learns, and collaborates autonomously. Their emergence signals not only the end of traditional SaaS as we know it, but the dawn of an intelligent, adaptive, and industry-aware software economy.

Agentic Software: The Successor to SaaS

As we step into a new era in enterprise technology, a transition from SaaS to agentic software that is as foundational as the shift from packaged software to the cloud. Where SaaS required users to navigate predefined interfaces and structured workflows, agentic software adapts dynamically to user intent, executing entire processes end-to-end with minimal oversight. These systems act less like tools and more like autonomous software employees, reasoning through ambiguity, learning from interaction, and optimizing outcomes over time.

This emerging paradigm redefines the software category itself. Agentic software, especially in vertical applications, is not merely a new feature layer; it’s a new economic architecture, combining software, services, and digital labor. These agents can operate continuously, scalably, and with greater precision than human teams, ushering in a profound rethink of how business gets done.

Agentic software is beginning to hollow out traditional SaaS systems. In industries such as automotive retail, legacy platforms built on static databases with layered UX/UI are being bypassed altogether. Rather than relying on rigid dashboards, AI agents now interact directly with the underlying data infrastructure, executing tasks intelligently and seamlessly without user mediation. This shift marks a fundamental departure from the UI-bound paradigms that defined the last two decades of enterprise software.

Where AI once functioned as a co-pilot, supporting human operators, it is rapidly evolving into autonomous agents capable of independent action. These systems now scrape competitor websites, analyze inventory in real time, and respond to customer queries across text, voice, and email, all without human supervision. They are tireless, accurate, and unburdened by fatigue or inconsistency, ideal for replacing roles that are repetitive, process-heavy, and prone to error.

The earliest and most visible disruption is occurring in marketing. Dealer operations, for instance, are seeing their tech stacks reconfigured by AI. In a sector notorious for churning through tools, AI agents have carved out a foothold in lead scoring, ranking prospects using behavioral and contextual data to enhance follow-up precision and increase conversion rates. What begins in marketing often cascades elsewhere.

Business Development Centers (BDCs) may be next. Historically costly and plagued by high turnover, BDCs are especially vulnerable to intelligent automation. AI agents are already capable of managing follow-ups, personalizing outreach, and executing omnichannel communications strategies with a level of consistency and scalability human teams struggle to match. The economic rationale for maintaining large, manual BDCs becomes increasingly tenuous.

This transformation poses an existential threat to traditional SaaS vendors. The prevailing model, seat-based pricing tied to static user interfaces and fragmented workflows, is collapsing under its own weight. As echoed in the Romance of the Three Kingdoms, “Empires long united must divide; long divided must unite.” SaaS once thrived by unbundling workflows into discrete tools. Now, AI is rebundling them into unified, intelligent systems that operate across verticals and eliminate the need for manual stitching.

Agentic software succeeds because it removes the traditional constraints of the SaaS model. Without the need for user-facing interfaces, these systems automate tasks more fluidly and without the bottlenecks of rigid hierarchies or dashboard dependencies. Where SaaS tools often go underutilized, businesses using only a sliver of available functionality, agentic systems deliver focused, high-ROI outcomes at lower operational costs.

Scalability is another key advantage. AI agents are always-on, require no additional cost per user, and can execute a large volume of tasks in parallel. This is especially compelling for small and medium-sized enterprises that cannot afford bloated software stacks or large teams. Agents offer speed, precision, and adaptability without requiring manual upkeep or extensive training.

The broader trend is clear: SaaS unbundled workflows; AI is rebundling them. What once demanded five separate tools and ten manual steps can now be accomplished end-to-end by a single intelligent agent operating through a unified data layer. Even transactions, previously gated by compliance and human intervention, are going autonomous, Visa’s recent move to allow agents to initiate payments underscores this evolution.

Importantly, this is not about adding AI to old systems. It is about building new systems AI-first. The next wave of startups will not retrofit intelligence onto SaaS, they will build from the ground up, with native AI capabilities, dynamic workflows, and continuous learning as their foundation.

The emergence of vertical AI agents marks a turning point in enterprise software, one that builds on the exponential momentum of generative AI, foundation models, and automation. This new category of intelligent software is rapidly gaining traction due to several converging technological and market forces that are accelerating adoption.

One of the most significant catalysts was the release of ChatGPT in late 2022, which radically increased public awareness and corporate curiosity around AI. The resulting “tremendous sucking sound in the market for AI” has made it easier than ever for new agent-based solutions to gain adoption, far faster than previous software paradigms like SaaS or mobile.

This acceleration is fueled by the relentless improvement of large language models (LLMs). Unlike traditional SaaS software, which improves in slow, incremental cycles, products built on top of modern LLMs benefit from the underlying model's rapid evolution. As new, more powerful base models are released, every product and agent built atop them “automatically becomes better”, a pace of innovation many in the industry describe as “insane.”

Yet LLMs alone aren’t enough. The true foundation of AI agent effectiveness is data. Enterprises are beginning to realize that to unlock real AI value, they must invest in organizing and enriching their proprietary data, cleaning it, structuring it into knowledge graphs, embedding it into vector databases. As one expert noted, “The foundation work for all AI is data.” In this context, a company’s internal data becomes a durable competitive advantage and a catapult to success.

At the frontier of this evolution lies the concept of agent swarms, networks of interoperable AI agents that collaborate and distribute tasks dynamically. This vision extends into what some call the agent economy, where AI agents don’t just pass information, they perform transactions, manage resources, assess trustworthiness, and interact as autonomous economic actors. For this to materialize, robust payment infrastructure is key. New systems are emerging to authorize agents to make payments, enforce spending limits, and act on behalf of their users in financial ecosystems.

The driving force behind vertical AI agents is their deep domain specialization. Unlike general-purpose tools, these agents are built to understand the nuances of specific industries and roles, whether it’s legal reasoning, handling patient records, debugging gameplay, or conducting enterprise-grade research. This domain expertise allows agents to automate point tasks previously done by humans: customer support, legal reviews, user interviews, data entry, and more.

Importantly, vertical agents scale effortlessly. They’re always on, can handle thousands of simultaneous requests, and cost a fraction of what equivalent human labor would. As they interact with users and accumulate data, they continuously improve, offering faster responses and more accurate insights.

In the bigger picture, vertical AI agents are reversing a decade-long software trend. SaaS historically unbundled workflows into specialized, often siloed applications. AI agents, by contrast, are rebundling these workflows, seamlessly orchestrating tasks across tools, data sources, and systems without dashboards, drop-downs, or rigid logic layers. This rebundling doesn’t just improve efficiency; it renders many legacy SaaS interfaces obsolete.

In short, vertical AI agents are not just a new interface, they represent a new operating system for enterprise work. Their rise signals the end of the traditional SaaS era and the beginning of a more fluid, intelligent, and autonomous digital economy.