What is Agentic AI & Why Should You Care About It?

The latest and greatest buzzword salad that dropped in the second half of 2024 is agentic AI, and while it might seem like hype now, it does have promising potential.

Metaphorically, agentic AI is a well-read, fast moving toddler who is borderline a savant.

Setting that half joke aside, it's a domain within artificial intelligence that:

•  Automates tasks fully or partially, on a scheduled or reactive basis
• Integrates with systems directly (e.g. via APIs or reading/writing data to databases)
• Interacts with desktop applications (e.g. web browsers, Excel) by impersonating human behaviors
• Adapts to changes and makes decisions based on context and reasoning, resembling human intuition

Based on publicly available information, agentic AI ranges from unpolished diamonds to plain old gravel:

•  Google's Project Mariner and Anthropic's "computer use" feature in Claude are both examples of agentic AI with guardrails, as both are operating autonomously to do things to achieve a user's request
• Salesforce's Agentforce is not agentic AI despite their marketing materials. Stripping away the buzzwords, that product boils down to prompt engineering and chaining (i.e. tweaking a prompt to accept inputs and produce usable outputs, then chaining multiple steps together)

So, given there's a murky marketplace, why should you care about agentic AI?

Let's revisit the the last part of our definition: it resembles human intuition.

If your organization has teams that regularly repeat the same set of tasks or business processes, especially with external stakeholders, the human intuition piece is a game changer.

Consider a few examples:

1. A financial services firm needs to open new accounts for clients using a custodian's web portal and digital forms
2. A nonprofit organization needs to reference and enter data in a local or state government website that was built in the late 2000s
3. An insurance tech startup is trying to incorporate compassion and empathy into its claims review process (which sounds sketchy, but hey, maybe it's the next Uber...)

All of those have a mix of repetitive work that relies to some degree on human intuition, especially based on the need to interact with an external stakeholder's data or technology stack.

Is it worth integrating agentic AI into your organization's systems stack?

Probably not yet, and we'll explore why tomorrow.

TLDR: Agentic AI is an exciting evolution that looks like true autonomy, although it's bleeding edge and not yet ready for primetime in your organization.