Can AI Agents Make Financial Decisions for You? What's Really Happening with Agentic AI Systems | Part 2 with Kelly

Should AI be allowed to move your money without asking first? It's not a hypothetical question anymore—agentic AI systems are already operating inside financial institutions, and consumer-facing applications aren't far behind.

In Part 2 of our conversation with Kelly, we dive deep into the world of agentic AI systems in financial services. These aren't simple chatbots or standalone AI models—they're sophisticated hybrid platforms combining large language models, machine learning algorithms, and autonomous software agents that can pull data, analyze patterns, and take action in real-time.

Kelly breaks down what makes these systems fundamentally different and why financial institutions are deploying them aggressively for internal use (particularly fraud detection and deep fake mitigation) while taking a much more cautious approach with consumer-facing applications.

EPISODE TIMESTAMPS:

[00:21] Introduction: What are agentic AI systems?

[00:28] Breaking down the technology: Why we say "agentic AI systems" not just "AI agents"

[00:40] The architecture: Multiple elements, software agents, and different tasks

[00:58] Beyond financial services: Where else agentic systems are deployed

[01:04] The role of large language models in orchestration and user interface

[01:14] Machine learning for deep quantitative analytics

[01:21] Why hybrid systems can do things individual models can't

[01:37] The autonomy difference: From reactive to dynamic real-time action

[01:52] Fraud detection use case: Fighting deep fakes and emerging threats

[02:08] Appropriate human oversight: Finding the right balance

[02:15] Personal financial assistants: The "pocket advisor" vision

[02:34] Beyond advice: AI executing transactions and managing daily finances

[02:45] The trust question: How reliable are these systems?

[02:57] Current deployment reality: Building with heavy oversight first

[03:17] The exciting potential: Greater dynamism and active support

[03:45] Where we're seeing adoption: Internal vs. consumer-facing

[03:58] Direct-to-consumer caution: Why banks are moving slower

[04:08] Shopping agents: E-commerce AI and the payment connection

[04:31] The "final yes" question: Why consumers still click to buy

[04:46] The future: Computer-to-computer interactions and changing parameters

[20:27] Credit underwriting deep dive: Data quality concerns

[20:39] Transparency and explainability challenges in AI models

[20:55] Comparing systems: Rules-based vs. machine learning vs. human judgment

[21:04] The human "black box": Subjective decisions are hard to pinpoint too

[21:18] Pros and cons: What machine learning improves (and what it doesn't)

[21:46] When things go wrong: Impact on customers, lenders, and the economy

[22:03] Human biases vs. human intuition: The complicated trade-offs

[22:18] Mission-based lending: Working with underserved borrowers

[22:37] High-touch meets high-tech: Marrying traditional relationships with AI

[22:52] The leverage effect: Data and technology expanding credit access

[23:09] Important limitations: What AI alone can't solve

[23:39] Processing benefits: Speed, efficiency, and consistency

[23:56] Machine learning nuances: Capturing patterns simple models miss

[24:10] Reliability concerns: Sensitivity to data changes

KEY TAKEAWAYS:

✅ Agentic AI systems combine LLMs, ML models, and software agents for autonomous action
 ✅ Internal fraud detection deployment is moving fast; consumer apps more cautiously
 ✅ Shopping AI exists now, but humans s