Vista Equity Partners released its 2026 report on Agentic AI, which argues that investors have been too quick to dismiss legacy enterprise software over fears of new AI-native competition.
“Software is not being replaced but is instead evolving into something more powerful,” writes Vista.
Join the more than 60,000 industry professionals who rely on our newsletter
As evidence, the firm references patterns from previous platform shifts. Most notably, the period from 2014 to 2020, when SaaS multiples compressed as investors debated whether AWS and other cloud platforms would displace incumbent vendors. Instead, software demand expanded.
Vista believes AI is likely to follow a similar pattern and argues that if AI adoption moves deeper into enterprise workflows, the application layer is still on track to capture the largest share of value.
It’s a different conclusion than recent doom-and-gloom projections that believe horizontal model providers like OpenAI and Anthropic will out-compete vertical vendors via ever-more-powerful general intelligence.
Corporate investment in AI across providers, infrastructure, and software tripled in a year to $37 billion in 2025, with the application layer taking more than half of that spend. Yet Vista also says that only 1 percent of enterprise data is incorporated into the AI tools workers use, even though 20 to 40 percent of workers already use AI in some capacity on the job.
That gap suggests adoption has started at the user level but not yet at the system level.
Lagging adoption may be partially blamed on the distinction between probabilistic models and deterministic systems. The variable nature of LLMs can be fine for use cases like customer service chatbots, but untenable for higher-stakes enterprise workflows requiring greater rigidity and accuracy.
That’s part of the reason why Vista is more optimistic on value accrual to the application layer. Per the report, application layer “agentic Enterprise Solutions combine both: AI-generated intelligence acting within governed, deterministic systems.”
Another source of optimism is the potential for revenue model upside. Once software can execute work rather than simply support an employee performing it, companies can move from seat-based licenses to charging for work completed.
Instead of being capped by headcount, revenue growth is then free to scale with volume (e.g. number of claims processed, incidents resolved, or approvals completed).
“This is not a story about replacing workers,” notes the report. “It is about software participating directly in specific, repeatable units of work where execution can be standardized, governed, and priced.”
Companies that can make this transition, says Vista, could eventually command higher multiples than traditional SaaS vendors.
Broadly, opportunity selection should focus on three key defensibility tests: context, trust, and scale. Does the vendor have enough workflow depth and domain specificity to make AI useful in production? Can it keep data and intellectual property inside the customer’s environment and deliver auditable decisions? Can it run reliably at enterprise volumes across geographies and business units?
On the other hand, vendors with generic workflows, limited proprietary data, and shallow customer relationships are facing real displacement risk.