The Rise of AI Agents: What Directors Need to Know

• Agents don't just advise—they act, triggering real processes like refunds, emails, or system updates.

• Human oversight is critical—for now—but agents are quickly moving toward autonomy, personalisation, and full enterprise integration.

• Directors must prepare now or risk falling behind in the most important software shift since the internet.

If you haven't heard about "agents" already, I'll be surprised—I think we'll remember 2025 as "the year of agents".

As happened previously with the internet wave, major software vendors like Salesforce and ServiceNow are busy re-organising their products around the generative AI wave—and more specifically around "agents" and the concept of a "digital workforce".

So what exactly are these agents, and how are they different from the generative AI tools we've been using for the past couple of years?

For a deeper understanding of the broader generative AI context, you can read my previous article on that topic.

Let's briefly review the evolution and current landscape of generative AI:

Phase 1: Chat / Copilots (2022)

The first true wave of generative AI adoption in business came through chatbots and copilots.

These tools excel at drafting documents, answering questions, summarising information, and coding. They're accessible through simple, human-like conversational interactions that anyone can use without technical expertise.

Products like ChatGPT, Microsoft Copilot, and GitHub Copilot demonstrated how AI could quickly assist in writing emails, generating code, and creating content.

Early adopters gained significant productivity advantages, with non-technical teams reporting time savings of several hours per day.

This phase can be characterised simply as: [prompt] → [response]

Phase 2: Reasoning

As AI models became more sophisticated, capabilities extended beyond simple text generation toward structured reasoning. These systems can now break down complex instructions into logical steps, follow multi-turn conversations, and solve layered problems.

The AI's "chain-of-thought" reasoning is often displayed in real time, improving transparency and reliability.

Reasoning broadened AI's application into more complex and abstract business areas, significantly enhancing tasks like research and notably reducing inaccuracies or "hallucinations".

This phase can be described (somewhat simplistically) as: [prompt] → [chain of thought] → [response]

Phase 3: Agents

Agents combine the capabilities of previous phases—but crucially, they add the ability to take action.

For example, with earlier chat-based or reasoning-based AI, a developer might receive suggestions for multiple code changes but would then manually apply these changes. By contrast, an agent could directly implement these code changes without further human prompting.

Initially, the types of actions that agents can perform have been deliberately limited, primarily for security reasons. For instance, in the coding example, an agent might be restricted to changing files within a single predefined folder, limiting potential risks.

However, as confidence grows, agents are progressively gaining broader access to organisational actions, including tasks such as:

• Sending emails on your behalf

• Updating customer records with relevant notes

• Raising refund requests within internal systems

• Draft-approving a recruit's application to study law

Given the significant implications, businesses are being advised to incorporate human oversight into agent-initiated workflows, particularly where substantial resources or money are at stake. For example, while an AI agent might initiate a refund request, a human supervisor should review and approve the final transaction. This cautious approach is strongly recommended to mitigate potential risks.

What's Coming Next?

The next phase of AI development will unlock further strategic opportunities:

• Autonomy: Agents will not only respond to direct prompts but will actively and independently look for tasks to complete. Initially, this might involve proactively identifying issues and opportunities, but eventually, agents could operate largely independent of explicit human instructions.

• World Modelling: AI will gain the ability to internalise representations of tasks or broader business goals. This capability will significantly enhance planning effectiveness, resilience in complex situations, and reduce errors.

• Retrieval Augmentation: AI's access to information will extend beyond immediate prompts to include internal databases and external sources. While current copilots already perform web searches, future agents will retrieve and integrate insights from a wide variety of organisational knowledge bases and internal systems.

• Personalisation: AI agents will become tailored to individual and departmental needs. They will remember and adapt to user preferences, work styles, workflows, and adhere strictly to personalised security rules.

• Enterprise-grade Security: Robust security measures will be essential to safely integrate agents into organisational operations. No business will grant unrestricted system access; thus, enterprise-grade safeguards will be vital to unlocking agents' strategic potential without compromising data integrity.

As directors and executives in Australian care organisations and charities, understanding and strategically deploying AI agents can significantly enhance your operational efficiency, reduce overheads, and amplify your organisation's community impact.

To strategically leverage these advancements, consider engaging with trusted advisors who can guide your organisation through adopting agents securely and effectively.

When you're ready, here's how I can help:

Andrew Walker
Consulting to for-purpose CEOs to deliver more impact with existing teams and systems - by freeing humans up from admin.
https://www.linkedin.com/in/andrew-walker-the-impatient-futurist/

Did someone forward this email to you? Want your own subscription? Head over here and sign yourself right up!

Back issues available here.