The Autonomous City: Agentic AI and Sell-Side Infrastructure
Sell-side transformation was never going to be easy. But the arrival of agentic AI as a serious enterprise technology has added a new wrinkle to many transformation roadmaps. Many multi-year programs didn’t originally account for operational functions that have begun automating themselves.
The impact has been twofold. First, it has already reshaped staffing models and sped up the urgency of data modernization. Second, it offers new tools and promises for solving some of transformation’s toughest legacy challenges.
New operators for the automated city
Even just a year ago, core sell-side functions like reconciling positions, triaging exceptions, and tracking settlement breaks took daily, labor-intensive work. Now, large sell-side institutions are finding ways to use AI to process transactions at higher volume, clear exceptions with little or no human intervention, and beat the clock on settlement cycles that had already accelerated to T+1.
Conversations about headcount are shifting in turn. For example, JPMorgan Chase’s latest Company Update was explicit about projections that AI will substantially reduce the operations workforce.
“The share of generative AI continues to grow as a percentage of our total AI activity. And overall, we’ve doubled the number of use cases in production this year. We’re focusing our efforts on the highest-impact areas such as customer service, including call center efficiency and personalized client insights as well as in technology, particularly for our software engineers.” — Jeremy Barnum, Chief Financial Officer, JPMorganChasei
Similar restructuring is mounting across the industry, with top banks such as Bank of America and Wells Fargo making explicit statements about their ability to redeploy or reduce workforces.
This shift creates a new challenge to the way firms build institutional knowledge. People used to enter the floor, book tickets, and spend years moving through the asset lifecycle, learning what happened at each step in equities, rates, structured products, or credit. Expert operators were the reliable resources who could diagnose a break at 4 PM when a clearinghouse pushed back an exception.
Now, the emphasis is shifting to AI fluency as a baseline hiring requirement. The thinking goes that if you bring in people who already understand agentic processes, you can train them on the asset lifecycle fundamentals that define things like trade capture, affirmation, settlement instructions, or the specifics of a structured product versus a rate instrument.
The operational infrastructure those programs are building, function by function, looks like urban planners trying to deal with aging water and power grids while planning for “smart city” analytics and the automation of resources.
Reaching the bedrock
Disruption aside, AI is also providing an unexpected advantage in addressing some of transformation’s stickiest and deepest problems. Banks carry core processing infrastructure that has been in production since the 1980s and 1990s, running critical functions that no modernization roadmap has been able to replace reliably. AI can help with those “untouchable” systems that seem too embedded, too expensive to analyze, and too risk-laden to replace.
Many of those bedrock systems run on mainframes or other legacy application architectures. That helps explain why sell-side firms continue to invest in legacy modernization. A 2025 Celent study found that 68% of firms plan to replace or significantly upgrade critical systems in 2025.ii A recent announcement from Anthropic showed how AI can upend the planning assumptions behind modernization, referring to the continued use of COBOL within banks. AI may accelerate transformation that previously seemed too challenging to undertake.
“COBOL modernization differs fundamentally from typical legacy code refactoring. You aren’t just updating familiar code to use better patterns, you’re reverse engineering business logic from systems built when Nixon was president. You’re untangling dependencies that evolved over decades, and translating institutional knowledge that now exists only in the code itself.” — Anthropiciii
For sell-side transformation programs now several years in, this shift offers a mechanism that fits alongside cloud migration work already in progress. Teams that spent the past decade designating some legacy constraints as a no-go area in their modernization strategy are revisiting those assumptions. It means that AI is potentially widening the scope of what transformation can achieve. The core systems that transformation programs spent a decade treating as fixed parts of their sub-structure are no longer out of scope.
What the autonomous city runs on
Fragmented source systems and different data definitions for core concepts like a customer, asset, or position produce AI outputs that inherit those inconsistencies. That disrupts the AI’s ability to function reliably for bank-wide compliance, regulatory reporting, funding decisions, or client-facing processes. Data readiness is a fundamental transformation challenge, but deploying AI widens that gap, turning it from a challenge into a source of new operational risk.
This is where AI deployment and broader transformation converge. The data ecosystem program transformation that programs have been building is now also the prerequisite for the AI strategy their organization is committed to delivering. In some cases, the promise of AI has proven more persuasive to business units that are skeptical of the operational efficiency case alone. Unit-level leaders now see that data readiness helps their AI programs move faster and get deployed in higher-stakes contexts instead of viewing it as a top-down mandate. AI helps create self-interest.
Not every AI use case needs fully normalized data to deliver value. Some do. Three questions help determine which is which before committing to a deployment timeline.
Is the data in this specific domain already clean?
If the relevant systems for a process or line of business have been through data modernization, AI readiness is high. But if the source data is still fragmented or inconsistent, data modernization must come first.
Is this an output someone reviews before acting on it?
Internal tools where experienced staff check results before making decisions can run while data work continues. Data impact is low. AI suitability grows up because humans will catch errors before they cause damage.
Does this output feed a process that is hard to correct?
Regulatory filings, funding calculations, and client-facing reports need reliable data. That raises the bar for AI transparency. The harder it is to unwind a wrong answer, the more you need to proceed with caution on both data and AI fronts.
Getting these questions right before setting a timeline is what separates AI programs that build on transformation work from those that generate new cleanup projects alongside it.
Building a city that keeps changing
At a small-group meeting of senior operational leaders in 2025, participants were asked what the next six months of agentic AI would produce in their organizations. They struggled to answer because it’s all happening so fast. That moment reflects why many transformations can’t see what’s coming over the horizon with AI. The rate of change is accelerating, and the next wave of transformation is already looming.
Although the nature and speed of that next wave are hard to predict, what’s clear is that more change is coming. Transformation doesn’t just reach a finish line. At the same time, innovation happens at the bleeding edge and then quickly becomes table stakes. Quantum computing, for example, may be on the same horizon where AI found itself three years ago.
Each of these waves of technology and business transformation faces an identical challenge: building something today that can absorb change when it arrives, while not having to start from the foundation again as needs and technologies evolve. Adaptive data ecosystems, modernized infrastructure on a cloud foundation, and developed workforces capable of learning new systems remain the best bet.
Whatever the next transformation cycle brings, transformation itself is the skill that sell-side firms must harness. More than any single initiative or completed milestone, it’s the source for competitive advantage in the decades ahead.
Authored By
Ted O’Connor
Ted is a Senior Vice President focused on Business Development at Arcesium. In this role, Ted works with leading financial institutions in the capital markets to optimize data, technology, and operational needs.
Share This post
[i] JPMorgan Chase, 2026. https://www.jpmorganchase.com/content/dam/jpmc/jpmorgan-chase-and-co/investor-relations/documents/2026-company-updates/company-update-full-event-transcript.pdf
[ii] Celent, 2025. https://www.celent.com/en/insights/dimensions-capital-markets-it-pressures-and-priorities
[iii] Anthropic, 2026. https://claude.com/blog/how-ai-helps-break-cost-barrier-cobol-modernization
