FTV’s Kapil Venkatachalam on Enterprise AI’s Reality Check

The modern AI stack is being built before our eyes. Critical components like core infrastructure and foundational large language models (LLMs) are still maturing. But when companies from Toast to Taco Bell to Bank of America are seeing millions added to the bottom line from AI experiments like personalized assistants, it’s clear the technology is beginning to transform how businesses operate.

While the builders of LLMs, as well as those manufacturing next-generation hardware, continue to plow through billions of dollars in outside investments, FTV is focusing our efforts on where AI will ultimately deliver value for businesses: the application layer.

Through our sourcing efforts and ongoing dialogue with executives in our Global Partner Network®, we regularly gauge how companies from startups to the Fortune 1000 are strategizing around AI. That helps us cut through the noise and uncover sustainable, long-term investment opportunities in our four sectors: financial technology and services, enterprise technology and services, vertical software, and healthcare technology and services. It’s led us to companies like Orbus Software, where AI aligns IT architectural governance with business goals; Windward, which uses AI to give visibility into global supply chains; Arden Insurance Services, which builds AI into its property underwriting process; and Kore.ai, which helps companies build end-to-end agentic AI experiences.

As AI races forward, we asked Kapil Venkatachalam, partner at FTV, for his thoughts on where enterprise AI is headed. Here, he digs into struggles in enterprise AI adoption, the shape of the market today and the real promise of agentic AI.

What’s the biggest gap you see between AI demos and real-world enterprise deployment?

The gulf between slick AI demos and actual enterprise adoption starts with infrastructure. At an AI startup, you can launch a pilot with clean data and perfect APIs. At a 50-year-old enterprise company, your AI pilot runs alongside ancient COBOL systems and data in 20 formats, so there’s a fundamental format and data readiness problem. Enterprise companies also deal with complex governance frameworks and a maddening tangle of red tape, which makes AI projects look risky. Combine all this with lack of AI talent, and a company may simply drop an AI project that didn’t achieve escape velocity or show any long-term results.

What will AI’s impact be in the coming years? Which industries will feel the effects first? 

Like most trends, the market is overestimating what can happen in the short term but underestimating the long-term potential. Still, the basic trajectory is unmistakable. OpenAI is seeing huge subscriber growth, IT jobs related to AI are surging, hardware providers are spending enormously on new AI data centers and AI developers are multiplying at a startling pace.

But just how substantial those gains will be is still up for debate. Many factors will determine economic impact, which will vary by industry. There’s no “golden path” to AI adoption. Instances where there are many iterative interactions between digital technology and the real world will take longer to be disrupted than knowledge work. It’s easier to show a quick AI win with email conversion rates than a quick win in, say, supplier negotiations.

No simple number will show the transformative effects of AI. Complex human decision-making has so many inputs that replicating it as a task will be quite the feat. If we get there, the gains could be tremendous. While sectors like technology, finance and professional services will likely see more immediate adoption and impact, all industries will ultimately be affected. The question is: how much?

There’s a lot of AI hype. What are the hard lessons enterprise companies are learning?  

In our interactions with current and prospective portfolio companies, we estimate 50% have launched at least one internal AI use case. This means the other 50% are still battling it out in PowerPoint. Companies must start somewhere, and the learning that happens through this journey is a critical first step. At the same time, we estimate that less than 20% of those current AI projects today make it to production.

Among enterprise giants, prevailing opinions about AI are changing. Leaders are wrestling with questions like “which use cases rise to the top of the priority list, and which justify the expense outlay?” More companies are joining Bank of America, which now has 90% of its employees using its internal AI agent. Or Klarna, which cut $10 million from its marketing budget with AI. As Raj Koneru, the founder of our portfolio company Kore.ai, pointed out recently, once large companies crack the code on prioritization and implementation, demand will explode.

What are the risks of building AI into business processes?

Remember that AI can be quietly embedded into software. Think of Gmail using text generation, or analytics dashboards with embedded machine learning functionality.

Over time this can mean customers end up in a black box relationship with vendors – they use a product but don’t fully understand the paths to any outcome. Automation plus AI can also cause silent failures, in which a customer doesn’t realize they’re missing a big goal or operating out of compliance.

How can companies battle back?

Companies can push back in multiple ways: conducting AI dependency audits or investing in continuous AI monitoring. They should establish transparency and accountability with AI vendors at the start, educating their staffs about AI capabilities. One other strategy: build AI projects with modular architecture, where it’s easy to swap in new components.

At FTV, we’re focused on companies building platforms that can take advantage of the innovation happening in the infrastructure layer and use model building blocks to support the next-generation of AI-powered applications. Domain expertise is critical – to be successful, these application providers need strong subject matter experts guiding development.

What is the promise of the AI agent?

We’re at the threshold of a new era for AI. All tech paradigms move from simpler, more rigid forms to more dynamic ones. But an AI agent is different in that it takes a leap into independent action-taking. It can do things it wasn’t explicitly programmed to do. An agent absorbs the context of a situation, sees a goal and breaks down the goal into a series of tasks, just like a human does. Then it learns from each task as it goes. It is not a chatbot, and not another AI tool.

The team at Kore.ai says Fortune 100 leaders sit up straight in their chairs when they see the diagram below, about orchestrating AI agents at scale. In it, AI agents handle predictable tasks, which frees up humans for creative, strategic work. A continuous feedback loop optimizes workflows on both sides. Humans play to their strengths, agents play to theirs. The system accomplishes much more than either agents or humans could alone.

The enterprise AI market will begin to mature as more agentic AI experiments move from pilot mode to production mode. That transition will give rise to a whole crop of companies that don’t yet exist, or perhaps do exist but were overlooked.