I built TARX, a personal assistant to help me with on-call support and operating our tech systems.
Since the start of this year, the AI space has exploded. New apps, new models, new tools, it's been relentless. Even as a developer, I can barely keep up. Tools like Claude Code and new models from major players keep raising the bar. Fast.
So I decided to build an AI assistant to help our small startup team operate our fintech systems without burning out. I've always loved the scene of Cooper and his robot TARS going on missions together. So I named him TARX. The first goal was simple, is to save time on daily tech operations and on-call support.
To set him up, I installed Claude Code on an EC2 instance running Debian Linux, giving him full CLI access to his own machine. I created a dedicated email and GitHub account, then added him to our code repos. I also set up a service account to give him access to our GKE cluster and selected Google Cloud services, read-only for almost everything, for safety. Before he started, I defined clear work conduct and ethics.
TARX has been remarkably efficient at on-call support. When an alert fires, he checks the logs, inspects the code, and delivers an analysis. If asked, he'll even suggest a fix and open a PR, which I then review myself. Especially useful when I'm traveling.
Code review is dead simple with TARX. Paste a link, and he's done. Better than most tools out there, with zero setup. And it's free.
I also connected him to our data warehouse. He can analyze business metrics and answer questions on demand.
Finally, I added him to Google Chat, where our team works every day. He picks up alerts and stays in the loop on what's happening.
Looking ahead: 2026 and beyond
2026 is shaping up to be the year of the AI personal assistant, one that works 24/7 without breaks. I haven't had time yet to make TARX fully autonomous with a heartbeat loop, but it's coming.
The autonomous team is real, and it's more accessible than ever. Behind it is an emerging agent economy, services that agents consume to do their work. Developers and startups are already rushing into this space. The closer we get to deep integration with company docs, data, and business context, the more this becomes a true AI employee which is cheaper, always available, and highly productive. Eventually, an entire team of autonomous AI employees.
My next idea: replicate TARX for my team members, and expand it to other companies in my network. AI employees that help companies move faster, automate repetitive work, and operate leaner.