How to Think About Building AI Agents in Your Organisation
There’s a lot of hype around AI agents right now. If you're reading this, you've probably seen the headlines, the demos, and the buzzwords. “Autonomous agents,” “multi-modal reasoning,” “end-to-end automation.”
It’s exciting. It’s overwhelming. And for many business leaders, it’s just another thing added to an already crowded list of priorities.
In fact, most senior leaders I speak with (especially outside of IT) tell me they’ve been handed some form of AI-related KPI for this year. Whether it’s exploring how AI could improve operations, building a prototype, or rolling out a pilot, there’s mounting pressure to do something with AI in 2025.
With all the noise, it’s easy to feel like you’re behind. Especially if your organisation is early in its AI journey, it can be tempting to jump straight into the deep end and start asking: “What kind of agent should we build?”
But here's the reality: Generative AI isn’t magic. It’s incredibly powerful, but it still needs to be thoughtfully applied. And jumping straight into building complex AI agents without the foundational experience or understanding can lead to wasted effort, missed opportunities, and probably some element of risk.
So Where Should You Start?
Even if your end goal is to build agentic AI into your organisation, I find that most teams benefit from hands-on experience with simpler, more practical applications first. This allows you to build capability, validate use cases, and lay the groundwork for more complex automation down the line.
Here’s a helpful progression I often recommend:
1. Start with a Q&A chatbot
Use whatever tools you have: Microsoft 365 Copilot, a custom chatbot built with Azure/OpenAI etc, or an off-the-shelf platform (there are plenty to choose from, if your data isn’t too sensitive). The goal is to give your people a faster way to access information buried in documents, intranets, or shared drives.
2. Enable source cross-checking
Extend the chatbot’s capabilities to cross-reference documents. For example, in an engineering context, upload a subcontractor’s report and compare it against your project scope or internal glossary. This is where you can begin layering in higher-order reasoning.
3. Review or augment manual, document-heavy processes
Start with low-risk areas. Maybe it’s reviewing legal documents, validating purchase orders, or analysing site reports. Let the AI assist with reviewing before you hand over any autonomy. Often, you’ll find efficiency gains without compromising control.
4. Use generative AI to analyse chat history
By understanding how employees are using your chatbot or AI assistant, you can identify high-value use cases that are already driving impact and build from there.
5. Use those insights to justify deeper investment in agents
When you're confident that a process is well-understood, high-impact, and repetitive, that’s when agentic AI starts to make sense. And by that point, you've already built most of what a quality agent needs: a clear workflow, defined inputs and outputs, and a proven ROI.
This staged approach also gives you time. Agent frameworks are still maturing and differ significantly in their capabilities. While the tech evolves, you can continue extracting value from simpler implementations without waiting for the perfect solution.
Case Study: An Australian Engineering Services Firm
I'm currently working with an engineering services firm that's exploring how to incorporate AI into their operations. Rather than jumping straight to building AI agents, we’re following a more measured approach:
We started by mapping their core activities across the organisation, focusing on tasks that are time-intensive and repeatable.
They’re providing a list of 20 high-effort tasks, along with example documents for each: drawings, emails, contracts, reports, etc.
Together, we’re shortlisting the top three based on impact, effort, and feasibility.
We’re starting with chatbots and light-touch workflow automation. For example, using AI to pre-fill reports or check contractor submissions against a scope.
Once we’ve proven those use cases and built confidence internally, we’ll look to add more autonomy through agents. But for now, the focus is on building experience and capturing early wins.
Final Thoughts
If you're unsure where to begin, look at your core business and hunt for inefficiencies. A good rule of thumb? The next time you open Word, PowerPoint, Excel or a PDF reader, ask yourself:
Why am I reading or creating this? What am I actually trying to figure out?
That question often reveals a task that AI can help accelerate or improve without needing to go to the full extent of building an agent from day one.
Building agents is exciting and we’ll be working with them every day soon enough. But it’s not the first step. It’s a milestone you work toward - and remember it’s just a technology, and for technology to be helpful it needs to first of all be useful. Focus on the problems that matter, use the tools you already have, and build from there.
Let me know if you'd like help identifying where to start in your business. Sometimes all it takes is one workflow to unlock momentum.