Citi and Hebbia on Agentic AI in Financial Services

At the AI Agent Conference in New York, a session featuring Sirisha Kadamalakalva, Managing Director and Global Head of AI/ML Investment Banking at Citi, and Aabhas Sharma, CTO of Hebbia, made a clear case: financial services is one of the most important enterprise AI battlegrounds right now — and most organizations are approaching it wrong.

The session covered why banking workflows are structurally well-suited for AI, what's actually blocking adoption, and why off-the-shelf AI deployments consistently fail in regulated financial environments.

Financial Services Moves Slowly for Good Reasons

The panel opened honestly. "Financial services has been slower to adopt AI. Security matters for all the right reasons."

This isn't organizational failure — it's appropriate caution in a high-stakes environment. The biggest blocker to AI adoption in banking isn't capability. The models exist. The workflows are clear candidates for automation. What's blocking deployment is trust, governance, compliance infrastructure, and the operational conservatism that regulated industries require. Those aren't problems you solve with a better model. They're problems you solve with better infrastructure and a longer time horizon.

Banking Workflows Are Built for AI Compression

Despite slow adoption, the panel was clear that financial workflows are structurally ideal for AI — precisely because they're so document-heavy, repetitive, and operationally expensive.

"There are thousands of documents. A lot of this work is mechanical."

Diligence, underwriting, research, deal analysis, financial modeling — these workflows involve analysts spending weeks reviewing documents, cross-referencing information, and manually synthesizing knowledge that AI systems can process in a fraction of the time. "Analysts often spend weeks on mechanical work. The agents are already picking up those tasks."

The opportunity isn't marginal productivity improvement. It's compressing weeks of analyst labor into hours, at scale, across the document-heavy workflows that define institutional finance.

The Analyst Role Is Shifting, Not Disappearing

This came up consistently across the session and it's worth being direct about: AI doesn't eliminate the financial analyst. It changes what the analyst does.

As AI handles document review, model preparation, and repetitive diligence, bankers shift toward the work that still requires human judgment — relationships, strategic thinking, negotiation, and decision-making. "People focus more on the creative aspects. The mechanical work gets automated."

Financial professionals increasingly become orchestrators of AI workflows rather than executors of them. That's a meaningful shift in skill requirements, and the firms that understand it early will train and hire accordingly.

Off-the-Shelf AI Breaks in Financial Workflows

One of the strongest sections of the session. The panel was direct about a failure mode playing out across the industry right now: organizations deploying generic AI solutions into regulated financial workflows and wondering why they don't hold up.

"You can't just use one model and hope it works. You need deterministic workflows. The single-shot approach breaks."

Financial AI requires multi-step reasoning, auditability, structured process control, and reliability — not because these are nice-to-haves but because the regulatory environment demands them. A single-model, single-prompt approach that works in a consumer context collapses under the requirements of investment banking or underwriting. "You need deterministic workflows" is not a technical preference. In finance, it's a compliance requirement.

Security and Trust Are Non-Negotiable

Banking operates on proprietary data, regulated information, and confidentiality obligations that have no equivalent in most other industries. The panel returned to this repeatedly. "These systems require trust. The data is highly sensitive."

This is why vertical AI vendors specializing in financial services are becoming strategic infrastructure partners rather than commodity software providers. Banks cannot move fast internally — the infrastructure complexity is enormous, the compliance requirements are specific, and the operational knowledge required to build trustworthy systems takes years to develop. "These workflows are extremely specialized. The infrastructure has to be built carefully."

The implication: generic AI vendors will not win regulated financial markets. Domain-specific platforms built around financial workflows, security requirements, and auditability will.

Organizational Change Is Harder Than Technical Deployment

The closing theme of the session was one that recurred throughout the conference in different forms. Technology is not the primary barrier to enterprise AI adoption in financial services. "The change management is hard. The workflows need to evolve."

Organizational adaptation, workflow redesign, trust building, and process integration are harder and slower than deploying the technology itself. The firms that treat AI adoption as a technology project will struggle. The ones that treat it as an organizational transformation — with the change management investment that requires — are the ones that will actually reach production at scale.

"We're still early. There's enormous opportunity ahead." The window to build the operational foundation correctly is still open. It won't be indefinitely.

This session was presented at the AI Agent Conference 2026 in New York. Speakers represented Citi and Hebbia.