I spent part of this weekend building my first AI agent from the ground up. Not because AI is the latest hype cycle, but because understanding where a technology might actually add value requires hands-on work. Across a career spanning finance, operations, and software, one thing hasn’t changed. Technology moves fast, but real value takes time to surface.
I have been writing software since the mid-1990s. One lesson has remained constant through every technology wave: it is important to continually look beyond the boundaries of what you offer today and understand what is coming next.
As the owner of an independent NetSuite consultancy, most of my time is spent delivering practical solutions for companies already running on the platform. My clients are manufacturers, distributors, and operators who expect results. They are not interested in hype. They want systems that run better, faster, and more reliably. Because of that responsibility, I have been careful in how I approach the current wave of AI.
Like many developers, I have been watching closely, experimenting cautiously, and looking for the places where this technology can truly add value rather than noise.
The first signs appeared directly inside the NetSuite ecosystem. At some point I noticed a new SuiteScript library called N/llm. Out of curiosity, I wired it into a quick experimental Suitelet: a text field, a submit button, and a simple prompt. It worked. You could ask questions. The system would respond. But the results felt closer to a SuiteAnswers lookup than something operationally useful. It was slow. It appeared to be metered. And the responses were not particularly actionable. So I filed it away mentally as an early signal, but not yet something I could responsibly deploy into a production environment for clients.
That is an important distinction. Just because something is possible does not mean it belongs in a business workflow.
Where AI has already improved my day-to-day work is as a development companion. I subscribed to the OpenAI platform and began using ChatGPT to review and analyze code. Sometimes it is a snippet. Sometimes it is a full SuiteScript function. Occasionally it is an entire Suitelet paired with a client script. Used carefully, it has expanded the range of approaches I consider when solving problems. It can suggest patterns, optimizations, or alternative structures that I might not have reached as quickly on my own.
That said, it is not magic. There are still plenty of moments where I have to push back, correct assumptions, or explain why something is possible inside the NetSuite environment when the model believes it is not. In other words, AI is not replacing the developer. But it can absolutely accelerate a developer who understands the system. That alone has been valuable.
Last weekend I decided it was time to go one step further. Instead of only using AI as a tool, I wanted to understand how to build an AI agent from the ground up. My goal was simple: learn enough to determine where this technology can realistically create value inside the NetSuite world. I started a new ChatGPT project titled: “My First Agent.”
My opening prompt was straightforward: “I need to learn how to build my first AI agent. Something common and basic to build my confidence. I am ready to begin on a Mac Mini with a folder named AI_APPS at the root level of my iCloud Drive.”
What happened next surprised me.
Rather than jumping straight into code, ChatGPT walked me all the way back to the operating system level. It started in the Mac terminal. It had me confirm my environment, install development tools, configure the OpenAI SDK, and structure the project directory. Every step was explicit. At times it moved too quickly. My strategy became simple: pause the process, upload a screenshot of what I was seeing, and ask for clarification. Each time, it would not only fix the problem but explain why the step mattered.
That alone turned the process into a learning experience rather than just a recipe.
Early in the setup process it began pushing toward Node.js. The reasoning was straightforward: Node runs instantly on macOS and many tutorials start there. But I paused and asked a different question. "Honest opinion… isn't Python considered more applicable to industry in general?" The response was, if someone says "AI engineer," most of the time they mean Python.
Python dominates the AI ecosystem because it excels at data analysis, system automation, and software integration. I had seen this myself while working alongside some very talented engineers over the years. It also reminded me of something closer to home: my kids had used Python extensively in business school for analytics and data work. So we pivoted. The path moved toward Python, virtual environments, and VS Code. Interestingly, ChatGPT intentionally kept me out of the new Codex IDE. Instead it walked me through setting up a proper development environment so that I could isolate dependencies and avoid polluting the underlying system. That decision made sense. Before you build agents, you need to understand the plumbing.
Like any emerging technology, understanding where AI agents may fit requires observation, experimentation, and hands-on learning. The goal is to separate what is interesting from what is genuinely useful.
After nearly three decades in software, one pattern has remained constant. Technology evolves quickly. Real value takes longer to reveal itself.
The challenge is not learning every new tool. The challenge is recognizing when a technology has matured enough to meaningfully improve the systems people rely on every day.
Left Ledger is an independent NetSuite consultancy based in Pittsburgh.
We do not resell licenses.
We do not collect commissions.
And we do not push implementations.
Instead, we work with companies that already run NetSuite and want to get more out of the system they own. If your financials, CRM, operations, or reporting workflows could be operating more efficiently inside NetSuite, I am always happy to have a conversation.
Here's what we've been up to recently.
This article is relevant to you if you are running your business on Oracle NetSuite and would like to add an isolated custom note on a transaction line. This article demonstrates a simple end to end architecture to customize and extend the capabilities of NetSuite.
This article will be relevant for you if your company is using NetSuite ERP, uses standard inventory items and a new business requirement arises where you now need the functionality of Serialized or Matrix for an existing item.
From us to your inbox.
