The Audit Podcast: Ep 238 – The Best Advice from Agentic AI Experts
Host: Trent Russell
Date: April 29, 2025
Overview
This "best of" episode dives deep into agentic AI: What it means for internal auditors, its real-world applications, the current hype, and where auditors should focus their time and attention. Trent Russell brings together insights from leading AI experts—Charles King (KPMG), John Thompson (ex-EY, Hackett Group), and Andrew Clark (Monitar)—while also offering his own guidance. The conversation blends practical frameworks and critical thinking, aiming to demystify agentic AI for audit professionals and challenge both hype and misconceptions.
Main Discussion Points & Insights
1. What is Agentic AI?
- Definition & Difference from Existing Tools
- Agentic AI refers to tools that perform actions on behalf of users, not just providing answers but initiating and completing tasks.
- Unlike knowledge assistants (e.g., ChatGPT responding to queries), agentic AI "takes action for you"—like composing, formatting, and nearly sending an email, or prepping files and reports automatically.
(Charles King, 04:49)
- Simple vs. Intelligent vs. Polymorphic Agents
- Simple agents respond to prompts and may monitor environments to trigger actions.
- Intelligent agents mimic human-level capacity: "They can buy a car, they can spend money, they can bind you to a contract."
- Future agents (polymorphic) will "build themselves" out of available digital building blocks or create new ones as needed.
(John Thompson, 09:13)
- Core Framework (Perceive-Reason-Act-Feedback)
- The classic agentic AI cycle: perceive (gather data), reason (analyze), act (make/execute decisions), and feedback (learn and adapt).
- Caution: Many solutions rebranded as "agentic" are, in fact, just advanced automation.
(Andrew Clark, 12:35)
2. Why Should Auditors Care?
- Governance is Key
- The governance of agents and agentic AI is where auditors "definitely need to be involved" because of the risks and possibilities for significant automation.
(Trent Russell, 00:50)
- The governance of agents and agentic AI is where auditors "definitely need to be involved" because of the risks and possibilities for significant automation.
- Practical Audit Use Cases
- Real internal audit teams are already implementing agentic AI for tasks like SOX controls testing, bringing theoretical advances into everyday workflows.
- Auditors are advised to explore, test, and understand these tools proactively, ideally in a safe environment ("Go in, break stuff... definitely be okay with that").
(Trent Russell, 19:06)
3. Expert Perspectives and Contradictions
- Charles King (KPMG)
- Emphasizes agentic AI’s practical power in automating multi-step audit tasks, reducing manual work far beyond what legacy RPA/automation provided.
- Suggests even "simple" agents add substantial efficiency: "If you can just sort of say, 'go find this thing and bring it to me,' that's really powerful."
(Charles King, 04:49)
- John Thompson (ex-EY, Hackett Group)
- Predicts "agents everywhere" by 2025 and beyond, paralleling prior hype cycles (e.g., GenAI, LLMs).
- Frames simple agents as automated responders, while intelligent agents can replicate almost any digital human action; polymorphic agents take this further.
- Stresses the importance of differentiating between LLMs and agents: "You do hear from people that, hey, I can't differentiate ... that's an issue for all of us."
(John Thompson, 09:13)
- Andrew Clark (Monitar)
- Provides a critical, "contrarian" take: agentic AI as a concept isn't new—it's just the latest iteration of automation, and marketing is running wild.
- Warns the term has become muddled: "The term agentic means absolutely nothing. It just means using a computer program at this point."
- Uses Google Maps as a classic (old) example—perceives input, reasons (route choice), acts (directions), feedback (updates with traffic).
(Andrew Clark, 12:35-15:21) - Cautions against replacing rigorous, proven operations research with LLM-driven reasoning just for the sake of buzzword compliance.
- Trent Russell (Host)
- Reiterates the real deployment of agents in audit: "Multiple internal audit teams have developed and rolled out agents within their internal audit department."
- Advises auditors not to fear experimentation: "Go in, break stuff in a safe environment and definitely be okay with that."
- Foresees an explosion of agentic AI in 2025 and urges listeners to engage critically rather than get swept up by hype.
Notable Quotes & Memorable Moments
-
On the transformative promise:
"Agents fundamentally, unlike knowledge assistants, ... are actually going and taking action for you. And they're really powerful and really flexible."
— Charles King (04:49) -
On practical impact:
"If you think about a lot of the actions that we take, regardless of where you are, ... if you can just sort of say, 'go find this thing and bring it to me.' That's really powerful."
— Charles King (06:27) -
On hype cycles and future trends:
"You're going to hear ... nothing but agents in 2025, no matter where you go. ... Just like it was all Gen AI ... it's going to be agents for the next two years."
— John Thompson (09:13) -
On the agentic AI framework:
"The general definition we have is really—it's four steps, which is perceive, gather data ... then there's a reasoning component ... you are now acting on those inputs ... and then there's a feedback loop."
— Andrew Clark (12:55) -
On critical skepticism:
"The term agentic means absolutely nothing. It just means using a computer program ... at this point."
— Andrew Clark (14:15) -
On practical advice for auditors:
"Find one of these frameworks you can play around with safely and break them ... Go in, break stuff in a safe environment and definitely be okay with that."
— Trent Russell (19:06)
Timestamps for Key Segments
- 00:00–04:49 – Trent Russell: Introduction, context for agentic AI, governance resource recommendations, and SOX audit use cases.
- 04:49–07:52 – Charles King: Simple explanation of agentic AI, difference from traditional automation, power for internal audit functions.
- 09:13–11:28 – John Thompson: Current/future state of agents, taxonomy (simple, intelligent, polymorphic agents), importance for audit.
- 12:35–19:06 – Andrew Clark: Contrarian critique, proper definitions, Google Maps analogy, marketing vs. reality, industry hype.
- 19:06–21:20 – Trent Russell: Personal advice, encouraging hands-on exploration, realistic vs. hyped perspectives.
Final Takeaways
- Agentic AI is both real and hyped: It's already shaping auditing, but much of the buzz stems from rebranding older automation concepts.
- Critical engagement is essential: Auditors should self-educate (using recommended guides and experimentation), distinguish between true advances and marketing, and focus on governance and risk.
- 2025 will see massive proliferation: Prepare to see "agents everywhere"—but remember, not every use deserves the label.
Practical Next Step:
Check out the AI Agent Governance: A Field Guide (Institute for AI Policy and Strategy, linked in show notes) for a structured deep dive and vet each "agentic" solution critically before investing time or resources.
