Building Support Agent: Bringing an In-House AI Assistant to Life
Every IT team reaches a point where the same questions keep landing in the same inbox. Where is the leave policy. How do I connect to the shared drive. Which form do I need to request new hardware. On their own these are small. Together they quietly absorb the hours that should be going toward the projects that actually move the business forward.
That recurring friction is what pushed me to build Support Agent, an in-house AI assistant for our team at GR Media Solutions. The goal was straightforward. Give people fast, reliable answers to everyday questions, and give the technical team back the time to focus on higher value work.
Why build it in-house
There are plenty of off-the-shelf assistants on the market, so the first question I had to answer was why build our own. The honest answer came down to control. Our processes, our policies, and our internal systems are specific to how we operate, and a generic tool simply would not understand any of that context. Keeping Support Agent in-house let us ground it in our own knowledge base and keep sensitive information inside our own environment rather than handing it to a third party.
How it came together
I started with the questions people were actually asking, not the questions I assumed they had. Pulling from real requests meant Support Agent was useful from day one instead of being a clever demo that no one needed. From there it was a matter of connecting it to the right sources of truth, testing the answers it gave, and tightening the gaps where it got things wrong.
The work was iterative. Every round of feedback surfaced something new, whether it was a policy that needed clearer wording or a system the assistant should have known about. Treating it as a living product rather than a one-time project made all the difference.
What I learned along the way
- Start with real problems. The most valuable answers came from the questions people were already repeating, not the ones I imagined they would ask.
- Keep a human in the loop. An assistant earns trust by being right consistently, so reviewing and correcting its responses early was time well spent.
- Good data beats clever prompts. The quality of any assistant is only as strong as the information it can draw on, so keeping our internal knowledge clean and current mattered more than any single technical trick.
What comes next
Support Agent is already saving the team time on the routine questions, but the real opportunity is broader. As we connect it to more of our internal systems, it can move from answering questions to actively helping people get work done. For me, that is the most exciting part. A tool that started as a way to reduce repetitive requests is turning into a genuine part of how the team operates.
Building it taught me that the best internal tools are not the most complex ones. They are the ones that quietly remove friction so people can focus on the work that matters.