The Rise of Vertical AI Agents. But First, Data.

Y Combinator is predicting that "vertical AI agents could mint 300 billion-dollar companies." At the same time, Alphabet plans to spend $85 billion on AI data centers this year.

Meanwhile, while everyone talks about building AI applications, notable investment activity is happening around data infrastructure. Snowflake just raised its annual revenue forecast to $4.4 billion, sending shares up 20% on AI demand.

While AI adoption accelerates, companies like Snowflake benefit from enterprises upgrading on modern data platforms. This might signal that enterprise companies are discovering they need to solve some foundational data problems and readiness before AI agents can better deliver on their promises.

The Enterprise AI Hypothesis: From Tools to Outcomes

Enterprise vertical AI agents could represent a shift in how business users interact with software. Instead of navigating multiple enterprise systems to complete a task, users might describe what they need done, and an agent could handle the entire workflow across those systems. Legal research that takes days might happen in minutes. Medical documentation that consumes hours could get completed during the patient visit.

If this works at scale, it's not just productivity improvement; it's a potential business model transformation. Bret Taylor, CEO of Sierra.ai, frequently talks about selling completed outcomes instead of software licenses. Success could be measured by tasks completed and time saved rather than adoption and engagement.

Y Combinator's Lightcone partners suggest vertical AI agents could become applications that "close the loop" from intake to action to audit. Their hypothesis: if AI can handle complete workflows in specialized enterprise domains, the market opportunity might dwarf traditional SaaS.

For enterprise users, this could change software expectations entirely! The interface might become conversational. The metric might become actual task completion, not clicks or someone logging into a platform (usage).

The Enterprise Reality Check: Data Determines Everything

Here's what the enterprise AI hype misses: vertical AI agents are only as reliable as the enterprise data foundation underneath them. An agent that drafts legal briefs needs access to current case law, firm precedents, and client matter details from multiple enterprise systems. An agent that processes insurance claims requires policy databases, medical records, and regulatory frameworks that live across decades-old enterprise infrastructure.

Most enterprises aren't ready for this. Their data lives in silos, lacks proper governance, and wasn't designed for AI consumption. Enterprise data complexity creates agents that work in demos or ‘proof of concepts’ but fail in production. They hallucinate facts, miss context, or produce outputs that require so much human verification that the efficiency gains disappear.

This explains why Snowflake's revenue surge is being driven by AI demand. Enterprises realize they need to solve data infrastructure before they can deploy reliable agents. Similarly, AWS introduced Amazon Bedrock AgentCore, marking a major infrastructure move toward operationalizing AI agents on its new platform.

The user experience implications are significant for enterprise adoption. An agent that occasionally gets things wrong isn't just unreliable, it’s low-trust and unusable for high-stakes enterprise workflows. Users need to trust that the agent's outputs are accurate enough to act on without extensive verification.

Case Study: How Casetext Built Enterprise Trust Through Data

The best example of getting enterprise AI right is Casetext's CoCounsel, which Thomson Reuters acquired for $650 million in 2023. When founder Jake Heller saw early GPT-4 demos, he made a decisive call to pivot the entire company to focus on what ultimately became CoCounsel, an AI-powered online legal research platform.

But the technical breakthrough wasn't just using GPT-4 for legal work. It was building the enterprise data infrastructure that made lawyers trust the results. Casetext created what Heller calls a "test-driven prompt" approach, writing gold-standard answers for legal tasks and shipping only when accuracy met law firm standards.

The UX innovation was equally important for enterprise adoption. Instead of just providing answers, CoCounsel surfaces citations, shows reasoning steps, and links to source documents within existing legal workflows, and closes the loop on completing the task. Lawyers can verify the agent's work without starting from scratch. This builds the trust necessary for adoption in a risk-averse enterprise environment.

The business results validated the enterprise approach. CoCounsel didn't just speed up legal research; it changed how law firms budget for it. Instead of billing hours for research, they could focus on higher-value analysis and strategy.

Design Principles for Enterprise-Ready Agents

Building agents that enterprise users and IT departments can trust requires different design thinking than consumer software:

Decompose expertise into enterprise skills. Instead of trying to automate entire enterprise jobs, identify the 10-20 specific tasks that experts perform repeatedly within existing enterprise workflows. Each becomes a testable skill with clear inputs, outputs, and success criteria. I’ve previously written about how using a framework like JTBD can do that across any business (large or small).

Make reasoning transparent for enterprise users. Enterprise users need to understand how the agent reached its conclusion for compliance and liability reasons. Surface the data sources, reasoning steps, and confidence levels that inform each output.

Design for enterprise verification workflows. The best enterprise agents augment human judgment rather than replacing it. Build interfaces that make it easy for users to check, correct, and learn from agent outputs within their existing enterprise processes.

Enterprise instrumentation is critical. Traditional enterprise software tracks user adoption. Agent software needs to track accuracy, reasoning quality, and outcome success. These metrics inform both user experience improvements and enterprise value measurement.

A Practical Enterprise Playbook

For design and business leaders considering vertical AI agents in enterprise environments:

Start with one painful, measurable enterprise workflow. Choose something users already do repeatedly within existing enterprise systems, where success is easy to measure, and where faster completion creates clear business value. Claims processing, compliance checking, and content analysis are good enterprise starting points.

Audit your enterprise data readiness first. Before building agents, understand what enterprise data you have, where it lives across systems, and how current it is. Most enterprise agent failures trace back to data quality issues that could have been identified upfront.

Build enterprise trust first, efficiency second. Design the user experience around verification and transparency that meets enterprise compliance requirements. Enterprise users who trust agent outputs will gradually rely on them more. Users who don't trust them will abandon them entirely.

Measure enterprise outcomes, not just adoption. Traditional enterprise software metrics like daily active users matter less than task completion rates, accuracy scores, and time savings. Align your measurement strategy with the business value you're creating for enterprise buyers.

The Enterprise Stack That Wins

Vertical AI agents represent the next evolution of enterprise software, but success depends on solving the foundational enterprise data challenge first. The companies winning this transition understand that agents are the application layer, but enterprise data infrastructure is the control plane.

Snowflake's surge quantifies how much enterprises are investing in this foundation. Casetext's $650 million exit shows how that investment converts to enterprise business value when applied strategically.

The opportunity is real for enterprise applications, but it demands the discipline to build enterprise data foundations that agents can rely on, user experiences that build trust within enterprise workflows, and business models that align with enterprise outcome delivery.

Strategic thinking before tactical execution.

Subscribe to Design Shift for more conversations that help creative professionals grow into strategic leaders.

Go from AI overwhelmed to AI savvy professional

AI keeps coming up at work, but you still don't get it?

That's exactly why 1M+ professionals working at Google, Meta, and OpenAI read Superhuman AI daily.

Here's what you get:

• Daily AI news that matters for your career - Filtered from 1000s of sources so you know what affects your industry.

• Step-by-step tutorials you can use immediately - Real prompts and workflows that solve actual business problems.

• New AI tools tested and reviewed - We try everything to deliver tools that drive real results.

• All in just 3 minutes a day

Join 1M+ pros