For the past three years, the enterprise software world has been buzzing with the promise of AI, a sentiment amplified by a constant stream of venture capital funding into a mushrooming ecosystem of startups. Yet, for most businesses, the reality has been one of costly experiments with little to show for it, as a recent MIT survey revealed that a staggering 95% of enterprises are not yet seeing a meaningful return on their AI investments. To make sense of this disconnect between hype and reality, we sat down with Vijay Raina, a leading venture capital expert specializing in enterprise SaaS. We delved into the critical shifts that could finally unlock AI’s value in 2026, exploring the evolution of AI companies from product vendors to strategic implementers, the true nature of defensibility in an age of rapidly advancing models, and the intense battle for a share of increasingly concentrated corporate AI budgets. We also examined the breakthroughs needed to move AI agents from isolated experiments to mission-critical collaborators.
The article notes that 95% of enterprises currently lack a meaningful return on AI. Considering VCs have been optimistic for three years, what specific technological or strategic shifts will make 2026 the year enterprises finally see real value, and what metrics will define this success?
That 95% figure really captures the frustration many executives feel. The initial wave was about broad experimentation, treating large language models like a silver bullet for every problem. The strategic shift we’re seeing now is a move away from that chaos. Enterprises are realizing that random pilots with dozens of vendors create more problems than they solve. The focus for 2026 will be on depth, not breadth. Instead of just plugging into a generic API, we’ll see a concentration on custom models, meticulous fine-tuning for specific tasks, and a robust ecosystem of tools for evaluation, observability, and data sovereignty. Success won’t be measured by the number of AI tools deployed, but by concrete ROI. We’re talking about a clear path to generating three to five times the initial investment, whether through top-line growth or by strategically reallocating labor spend toward technologies that demonstrably multiply output.
Molly Alter suggests some AI product companies will pivot to becoming “generalist AI implementers.” Could you walk us through the step-by-step process of how a company might make this transition and what challenges they would face in shifting from a product to a service-oriented model?
That’s a fascinating and very real transition we’re starting to observe. It begins with a company developing a sharp, specific product—say, an AI agent for coding or customer support. They land their first few enterprise clients and, through the process of implementation, they become intimately familiar with that customer’s unique workflows. The key pivot happens when they leverage that deep knowledge. Instead of just selling more licenses, they essentially replicate the “forward-deployed engineer” model, using their own expert team to build additional, custom AI use cases on top of their core platform. They transform from a product seller into a trusted AI consultant. The main challenge is that it fundamentally changes the business model. You’re moving from a scalable, low-touch software product to a high-touch, people-intensive service. This impacts everything: your margins, your hiring profile, your sales cycle, and your ability to scale rapidly. It’s a difficult tightrope to walk.
Jake Flomenberg questions moats built on model performance, asking if a company could survive a competitor with a 10x better model. What are the key components of a durable “workflow moat,” and can you provide an anecdote of how one is built in a specialized industry like manufacturing?
That “10x better model” question is one I ask founders constantly because it cuts to the core of defensibility. A model advantage can erode in months. A true, durable moat is built on embedding your solution so deeply into a customer’s operations that ripping it out would be prohibitively painful. This “workflow moat” has a few key components: it solves a mission-critical problem where failure breaks a production process, it accumulates proprietary context and data that is difficult to recreate, and it becomes the foundational layer for how a business operates. In manufacturing, imagine a startup that digitizes employee-led production processes. It starts by mapping a single assembly line. Soon, it becomes the system of record for the entire factory floor, transforming how they operate and building up proprietary data on efficiency and quality. Even if a competitor launches a slightly better AI for predictive maintenance, it doesn’t matter. The first company isn’t just a tool; it’s the operational backbone. Switching would mean halting production and retraining the entire workforce. That’s a workflow moat.
Several VCs predict that while overall AI budgets will grow, spending will become highly concentrated. What concrete proof points or ROI calculations must a startup present to a CIO to ensure they become part of this concentrated spend rather than being cut during vendor rationalization?
The era of experimental AI spending is closing. CIOs are tired of the vendor sprawl and are actively looking to rationalize their toolsets. To be one of the few vendors that survive and capture a larger share of the budget, a startup needs to present an airtight business case. First, the ROI must be undeniable and quantifiable, not just a vague promise of “efficiency.” You need to show how your solution saves time, reduces cost, or increases output in a way that holds up through rigorous procurement and security reviews. Second, you must prove you are mission-critical. This comes from having customers who are genuinely delighted and willing to take reference calls to talk honestly about your impact. A powerful signal is achieving an annual recurring revenue of $1 million to $2 million, but more important than the number itself is the story behind it—a story of how enterprises view your product as essential to their core business, not just a nice-to-have feature.
We see a split view on AI agents, from cautious optimism to predictions of them outnumbering human workers. What key technical and compliance breakthroughs are absolutely necessary for enterprises to move beyond siloed agent experiments and start deploying them at scale in mission-critical roles?
The vision of a workforce dominated by AI agents is exciting, but we’re still in the very early stages. Getting beyond siloed experiments requires solving some fundamental challenges. On the technical side, we desperately need standards for agent-to-agent communication. Right now, agents are isolated; an inbound sales agent has no context from a customer support agent. We need them to converge into a single, unified system with shared memory to enable truly contextual conversations. On the compliance side, the hurdles are immense. Enterprises need absolute certainty around agent governance. Before deploying an autonomous system that can interact with sensitive customer data or execute critical financial transactions, there must be robust controls, oversight, and a clear audit trail. This is less about making the agent smarter and more about making it trustworthy and secure within a regulated corporate environment. Without these breakthroughs, agents will remain helpful but limited assistants rather than core members of the workforce.
What is your forecast for the enterprise AI startup landscape over the next two years?
I believe the next two years will be defined by a great consolidation. The market will bifurcate sharply: a small number of startups that can prove undeniable, mission-critical value will capture a disproportionate share of enterprise budgets and see explosive growth. Many others, those who are just “nice-to-have” features or can’t prove their ROI, will see their revenue flatten or disappear entirely as CIOs rationalize their spending. We’ll also see a significant shift in focus from purely digital workflows to AI that reshapes the physical world—in manufacturing, climate monitoring, and supply chains. The winners in this new landscape will not be the companies with a temporary model advantage, but those who have painstakingly built deep workflow and data moats within specific, complex industries. Ultimately, in a market this crowded, the most critical differentiator will be the ability of founders to attract and retain top-tier talent to execute on their vision. Execution, more than anything, will separate the enduring companies from the vaporware.
