Fun and useful ways I use AI agents that keeps me happy

It is really difficult now to not think about, read about or work with AI. As someone working in the Software space, I can’t tell you how many times AI comes up. So rather than write yet another AI post trying to inform you of which model is best, I thought I’d share examples of how I use AI in a non-code-generation mode that makes my life easier.

System hardening

Once you have a project or product online, you are immediately exposed to the wild wild west called the internet. You have to deal with bad actors probing your system to find out if you’re running a vulnerable version of the software. The bulk of this seems to come from requests looking for WordPress or unprotected environment variables.

Using well-built software may have a lot of the security checks in place but you still have to deal with processing that request. One obvious solution is to place your service behind Cloudflare. Unfortunately some requests still get through.

I have been logging my requests to Grafana which is an observability product. I am able to read every request and see the user agent strings. Connecting AI to both the Grafana and Cloudflare MCP, I was able to scan and identify some bot traffic that pretends to be human and using the request pattern, update Cloudflare security rules to block such requests.

I am new to the Cloudflare stack. I can read the docs and probably learn to define rules, but reviewing the logs that come in and identifying when a user agent string is fake is something AI can do quickly.

In house Data Analyst

So, as stated earlier, I am pushing logs to Grafana. For product type data, I am using PostHog for capturing product activity. Part of my task is to review how well the product is doing from a user activity and workflow completion perspective. I can draw up or review a few reports on both platforms and try to see correlations and draw up some conclusions from it.

Here is the AI advantage. I connected it to the PostHog and Grafana MCPs and gave it the go-ahead to tease out some reports. It was also able to detect improvements or dips in metrics and what could have possibly caused them. I also told it to suggest experiments we can run based on the data we were looking at. It went on to generate a beautiful webpage with all the summary. This was a plus as I don’t do well with Excel.

User Interface Design

I have to confess, no matter how many times I try to get good at user interface design, I successfully fail at getting good at it. I have pretty much given up. If all I can do is add a drop shadow to a square, I will count that a win.

With Claude Design, I am able to brainstorm user experience flows and iterate on different ideas. Once I am satisfied with the result, I can then hand over the design to another agent to implement.

Before the age of AI, I would have to wait and beg some guy named Jake to collaborate on a project with me. I can’t tell you how many project ideas have been smashed against the rocks off the coast due to my lack of design skills.

So right now, I can get a project running and maybe when I have made enough money hire a UI/UX expert to make this better. At this point I will be paying, so — win-win for everyone.

Personal SysOps Engineer

I am a software engineer after all which means I am always tinkering with something. Many times I have had to employ AI to help diagnose and resolve system issues. A good one for example: I have an old MacBook Pro (2015) which is pretty much no longer supported by our dear Apple. I was however able to get Linux installed and running on it, but during one kernel upgrade, Linux could no longer work with the graphics card that came with the laptop.

After some probing and testing, I was finally able to disable one of the troublesome graphics kernel modules and get my dear old Mac going. So while Apple will be expecting me to shell out almost £3000, I am keeping this old boy going on Linux.

Next time you buy a laptop and you’re interested in Linux, shy away from Apple.

Do you have other non-code-generation uses for AI? Do share them; I would like to learn new tricks.