I’ve been building an invoice tracker with AI. When I call it, it retrieves my invoices from Romania’s National Agency for Fiscal Administration and shows me the ones I haven’t marked as paid. I tell it which invoices have been paid, and it updates the tracker so they don’t appear in the next call.
I value being able to build something useful for myself. Looking back at how I got here, I can also see how much I already knew when I started.
Knowing where to start
My expertise is on the business side, but I’ve spent years working with developers and hearing them discuss technical decisions. Some of that knowledge helped me find a way to build this tool.
I understood how APIs work and how systems can connect. My accounting platform didn’t offer an option for connecting to it, but I knew I could get the invoice data I needed through the tax authority’s API. That meant I could continue without depending on the platform.
Without that knowledge, I probably would have dropped the project. I might have seen the lack of a connection as the end of the idea.
I also knew to be careful with private and sensitive information. I knew I shouldn’t paste secrets into the chat or expose them elsewhere. And I knew not to hardcode passwords. Those were concerns I brought to the process from the beginning.
The point where I needed help
During the build, the AI assistant asked me to configure Google OAuth. I didn’t understand why that step was needed or how it connected to the rest of my project.
That was when I reached out to my technical colleagues.
They explained why I needed that type of authentication method and guided me through the login steps, explaining what was happening as we went. Their help gave me context for instructions I couldn’t assess on my own.
I’m still not fully sure what happens in the backend. Having a tool that responds as I expect hasn’t removed that uncertainty.
What my experience helped me recognise
Building this has given me more independence. It has also shown me how useful those years around developers have been. I had enough background to recognise another route to the data and to question a step I couldn’t explain.
Someone approaching the same project without that experience might have different questions, or might not know which questions to ask. That is easy to overlook when we talk about what a non-technical person can build with AI. We each bring different knowledge to the work.
For me, knowing when to ask a colleague was part of moving the project forward. Their explanation helped me continue and understand more of what I was doing.
What comes next
Before relying on the tool or suggesting it to a customer, I want a technical colleague to review the backend and data handling. I also want a QA specialist to test beyond the expected flow, including unexpected inputs and what happens when the tool fails.
I want to understand what data could be exposed, what needs fixing and what I may have missed. The help I received with the login setup doesn’t replace that broader review.
The findings will also help me plan the next phase: creating and sending invoices through conversation, without opening my accounting platform.
I want to keep exploring what I can build. AI has made it possible for me to take an idea further myself, and working with my colleagues has helped me move through the parts I couldn’t understand alone. The next step is to use that same combination of curiosity, experience and technical input to decide what the tool is ready for and how to develop it further.
