Executive Summary

DeepLearning.AI’s Agentic AI course by Andrew Ng provides practical training in building adaptive AI systems, but its minimal focus on ethics, despite examples like an AI autonomously deleting a project, underscores the growing risks in this field. With 30 years of hands-on experience in software engineering, data analytics, IT, and operational risk (primarily in Financial Services), I offer deep AI and IT risk management expertise to help Boards navigate this landscape effectively.

Why This Course?

As an expert actively using agentic AI in tools like Warp.dev to enhance coding, debugging, and even building custom AI agents, I review courses like this one (Agentic AI by DeepLearning.AI) to stay current, identify knowledge gaps, and evaluate what practices are being taught to emerging professionals.

Agentic AI refers to systems that leverage large language models (LLMs) to handle complex tasks by planning multi-step processes, executing them iteratively, and improving outputs through reflection and tool use. In other words, it's like a beehive where each bee (agent) has specialised jobs, coordinated in various ways with differing degrees of autonomy: What could possibly go wrong? ;) This setup amplifies efficiency but demands robust governance to prevent an unintended sting in the tail.

Even Andrew Ng acknowledges the hype:

"When I coined the term agentic AI to describe what I saw as an important and rapidly growing trend in how people were building LLM-based applications, what I did not realise was that a bunch of marketers would get hold of this term and use it as a sticker and put this on almost everything in sight. And that has caused hype on agentic AI to skyrocket."

Beyond the hype lies powerful tools that can be used for good or ill. Even in positive applications, risks emerge if not managed properly. The course excels in technical skills but addresses both ethics and risk management minimally, with brief mentions of guardrails, despite examples like an autonomous agent deleting a project?! This is surprising compared to other DeepLearning.AI courses. Validation and errors are framed technically, not ethically or through a risk lens, while balancing theory and in-depth practice. My experience is that this is typically lacking in most data science education.

Relevance for Boards

My extensive data analytics and AI experience aligns with the course’s hands-on focus. Boards can benefit from my first-hand ability to translate these tools into governance and risk management strategies, ensuring AI deployments are safe, compliant, and aligned with organisational objectives.

Risk Management Considerations

In Andrew's closing remarks for the course, the opportunities and risks are on display:

"[We've talked] about planning and multi-agent systems that can let you build much more powerful, although sometimes harder to control, and harder to predict in advance types of systems."

He continues:

"I want to thank you again for spending all this time with me, and I hope you will take these skills, use them responsibly, and just go build cool stuff."

Building and automating may be the exciting (and easier) parts for teams, but as Boards know, there's nothing new under the sun when it comes to overlooking risks in the rush to innovate!

Agentic AI introduces complexities through individual agents orchestrated by non-deterministic LLMs, creating significant potential for errors, inadequate testing, or validation failures, such as (so-called) hallucinations in high-stakes scenarios like fraud detection, especially under competitive pressures where speed and cost are often prioritised over thoroughness.

Enlightened Boards, particularly in regulated industries like Financial Services, should scrutinise how testing and validation are integrated into the development and deployment of these advanced systems to mitigate risks effectively.

The Bottom Line

My expertise in AI and operational risk management positions me to guide companies toward secure, innovative, and risk-aware AI adoption, helping Boards make informed decisions in this dynamic space.