From Generative to Agentic: How AI Is Rewriting the Enterprise Playbook

Artificial intelligence has moved quickly from novelty to workforce necessity. But as enterprise leaders look beyond experimentation, a new phase is taking shape—one that shifts AI from answering questions to actually doing the work of a human employee.

iCapital’s Nick Veronis, head of portfolio management, moderated a conversation with Monti Saroya, senior managing director and co-head of Vista Equity Partners’ flagship fund, and Michael Demissie, managing director and head of Applied AI at BNY. Together, they explored the rise of agentic AI, what it means for enterprise software, and why this transition is less about technology breakthroughs and more about trust, culture, and execution.

Why agentic AI is a different ball game

At its core, agentic AI represents an evolution of generative AI. While large language models excel at synthesizing information and responding to prompts, agents go further: they act.

As Saroya explained, agentic AI can be thought of as software users—digital workers that interact with enterprise systems the same way a human would. They log in, follow workflows, make decisions within guardrails, and execute tasks to achieve defined outcomes.

Demissie described how this concept is already live at BNY. The firm refers to its agents as “digital employees,” complete with logins, email addresses, KPIs, and governance structures. These agents aren’t theoretical or experimental; they operate in production environments, handling real workflows that once took significant human effort.

In other words, AI is no longer just a tool; it’s becoming labor.

Why enterprise moats still matter

With such rapid progress, it’s natural to ask whether smaller and more agile AI‑native startups will overtake established enterprise software providers. The panel’s answer was nuanced.

Saroya emphasized that enterprise moats are rooted in data, context, trust, and regulation. Mission‑critical systems from insurance platforms to financial infrastructure contain decades of proprietary data that have never been exposed to the public internet. That data can’t simply be scraped or replicated by a foundation model.

Equally important are security, auditability, and explainability. In regulated industries, every decision must be traceable. When a regulator asks how a conclusion was reached, “the model said so” is not an acceptable or legal answer. Enterprise software is built to satisfy those requirements, and that remains a powerful advantage as AI capabilities are layered on top.

Demissie echoed this from the perspective of a global financial institution. At BNY, scale and trust are not theoretical constructs but operational realities. Processing trillions of dollars in assets and payments requires absolute reliability. While that creates a strong starting position, it also raises the bar: clients increasingly expect the same speed, simplicity, and intelligence they experience elsewhere.

The message was clear. Incumbents can’t rely on their moats alone and they must modernize from within.

Cultural transformation is the hard part

Despite headlines about cutting‑edge models, both speakers agreed that technology is no longer the primary constraint, but adoption.

Demissie described this shift as a social contract between firms and employees. From the outset, BNY treated AI as a strategic imperative driven from the top. Nearly the entire workforce now has access to AI tools, with many already building custom applications.

Just as important, the firm has been transparent. Some tasks will be automated, and that reality isn’t hidden. Employees are encouraged to move toward higher‑value work as routine workflows are handed off to agents.

Saroya offered a parallel view from Vista’s portfolio, where adoption has varied even among engineers. Top performers tend to embrace AI quickly, while resistance clusters in the middle. The gap, he noted, isn’t technical skill, but belief.

Agents are expanding the market, not just efficiency

One of agentic AI’s most significant implications is economic. Traditional SaaS growth has been capped by seat counts, but agents remove that ceiling.

Unbound by headcount, agents enable usage‑ and outcome‑based pricing and unlock new markets—labor‑intensive services such as KYC, payments, research, and compliance once limited by human capacity. At BNY, Demissie noted that agents are already compressing workflows from weeks to hours, freeing teams for judgment‑driven work.

The software company of the future

Looking ahead, Saroya outlined how enterprise economics could shift. Traditional SaaS models depend on margins powered by human labor. Agentic models invert that structure: labor declines, while inference, the compute that enables agents to reason and act, becomes the primary cost.

The economics may look different, but margins can rise, growth can accelerate, and value creation can become more durable. The near‑term constraint isn’t software but infrastructure: compute, data centers, and power.

Would you put the genie back in the bottle?

Both panelists answered without hesitation: no.

Demissie framed AI as a societal superpower, capable of expanding access to healthcare, information, and expertise worldwide. Guardrails and discipline are essential, but the opportunity is too significant to ignore.

Saroya took a generational view. Knowledge work, he noted, is relatively new. If AI can absorb screen‑based tasks, people may be freed to focus on more meaningful problems.

Agentic AI, the panel concluded, isn’t about replacing humans. It’s about reshaping how work gets done, and for enterprises willing to lead that shift thoughtfully, the opportunity is enormous.