This is a periodic column by CAA Board Member Dan Alvarez, addressing technology issues in the banking world, for non-tech professionals.
#11 AI Agents PART TWO
(November 2025)
AI Agents: Your Future Digital Coworkers
(And Why They Won't Steal Your Lunch)
(And Why They Won't Steal Your Lunch)
Should I be worried about my job? Are AI agents going to replace people?
The honest answer: It's complicated, and it depends on what you do.
JPMC's consumer banking chief told investors that operations staff would fall by at least 10 percent in the next five years thanks to AI deployment. CEO Jamie Dimon has been candid that AI will eliminate some jobs, though he's also said the JPMC will retrain impacted workers.
Here's the pattern emerging: AI agents are best at automating repetitive, process-driven work.
Jobs most at risk are those involving:
- Rote processes (account setup, routine data entry, etc)
- Standard document creation
- First-level customer support
- Transaction processing.
Jobs that remain valuable (and may become even more so):
- Client relationship management
- Strategic decision-making
- Creative problem-solving
- Complex negotiations
- Jobs requiring empathy and emotional intelligence.
As Waldron explains, workers are shifting from being "makers" to "checkers" — managing AI agents rather than doing the work themselves. It's less about replacement and more about augmentation, though that's cold comfort if you're in a role that has been automated.
JPMC is also hiring aggressively in AI-related roles. It has already built a team of over 2,000 AI/ML experts and data scientists.
What are the risks? Can AI agents make mistakes?
Agents are powerful but far from perfect. It’s no secret that AI has made some very costly mistakes over the last few years.
Key risks include:
- Hallucinations: LLMs sometimes confidently generate plausible-sounding but completely false information. Imagine an AI agent creating a client presentation with made-up financial figures. Yikes.
- Security concerns: AI agents that can access multiple systems and take actions need incredibly tight security. One compromised agent could potentially cause havoc across connected systems.
- Bias: AI models learn from historical data, which means they can perpetuate or amplify existing biases in decision-making.
- Reliability: Unlike humans who can explain their reasoning, AI agents can sometimes produce unexpected results through opaque processes.
- Compliance nightmares: In a highly regulated industry like banking, ensuring AI agents comply with all relevant rules is no small feat.
This is why JPMC is taking a measured approach – starting with internal employee-facing tools before rolling out customer-facing applications. It's sort of like learning to walk before you can run.
How do AI agents actually learn and improve over time?
AI agents get better through a combination of techniques:
1. Continuous training: The underlying LLMs are regularly updated with new data. JPMorgan updates LLM Suite every eight weeks, feeding it more information from their databases and applications.
2. Fine-tuning: Models can be specialized for specific tasks or industries. JPMorgan is developing finance-specific models that understand banking terminology and workflows better than general-purpose AI.
3. Reinforcement Learning from Human Feedback (RLHF): When humans correct an agent's outputs or rate them, that feedback trains the model to make better decisions.
4. Integration Depth: The more systems an agent can access and the more context it has, the better it performs. This is why JPMC's strategy of connecting AI deeply to their proprietary data is so crucial.
Think of it like training a new employee: They start with general knowledge, get specialized training for your company, learn from feedback on their work, and gradually gain access to more tools and information as they prove themselves.
The key difference is scale. AI agents can learn from thousands of interactions simultaneously and share that learning instantly across the entire system.
What's next? Where is this technology heading?
Please note – the world of AI shifts so rapidly that my own analysis might be outdated in as little as a month.
Here’s where I think things are heading:
Near-term (1-2 years):
- More sophisticated agents that can handle increasingly complex, multi-step tasks
- Direct customer-facing applications (though with humans still in the loop)
- Integration across more business functions
- JPMC aims for 80% of applications to run on cloud platforms by 2025 to further accelerate AI deployment.
Medium-term (3-5 years):
- Agents that can collaborate with each other, forming "agent teams" to tackle complex projects across dozens of systems and data sources.
- More autonomous decision-making with less human oversight
- Personalized AI assistants for every employee and potentially every client
- Significant restructuring of job roles and organizational structures
- AI agents to handle operational tasks, such as triaging failures across tech stacks or remediating issues with internal systems.
Long-term (5+ years):
- JPMC's vision of becoming a "fully AI-connected enterprise" where every employee, process and client experience involves AI.
- AI agents that can proactively identify opportunities and problems before humans spot them
- Fundamental changes to how knowledge work gets done.
The question is which finance organizations will successfully navigate this transformation, and at what cost to their workforce.
As Jamie Dimon noted, AI's impact could be as transformational as some of the major technological inventions of the past several hundred years. That's not hyperbole when you're already seeing investment banking decks created in 30 seconds.
_________________________________________________________________
The Bottom Line
AI agents represent a fundamental shift in how work gets done. They're not magic, they're not perfect and they're definitely not going away. For us lay people, understanding this technology is crucial – whether you're still navigating your career in finance, leading teams through this transition or simply staying informed about where the industry is headed.
The firms that figure out how to blend human judgment with AI capability effectively will have a significant competitive advantage. JPMC is clearly making a massive bet that they'll be one of them.
Click for PART ONE
_______________________________________________________________
Links to Prior CAA Tech Corner Columns
#4 Anatomy of a Modern Credit Card
#5 How to Whitelist an Email Address
#6 Multi-Factor Authentification
#9 JPMC Meets Amazon in the Cloud
#10 Mobile Payments/Digital Wallets
About Dan Alvarez
Dan Alvarez began at JPMorganChase in June 2016 as a summer technology analyst/ infrastructure engineer, and left in April 2022 as a Senior Software Engineer in Global Technology Infrastructure - Product Strategy and Site Reliability Engineering (SRE). Since May 2022, he has worked for Amazon Web Services as an Enterprise Solutions Architect.
He is also an avid guest lecturer for the City University of New York and has given lectures on artificial intelligence, cloud computing and career progression. Dan also works closely with Amazon's Skills to Jobs team and the NY Tech Alliance with the goal of creating the most diverse, equitable and accessible tech ecosystem in the world.
A graduate of Brooklyn College, he is listed as an Alumni Champion of the school and was named one of Brooklyn College's 30 Under 30. He lives in Bensonhurst, Brooklyn.
----------------------------------
Comments? Questions?
Send them to chasealumtech@gmail.com.