AI Tools I Actually Use Every Day (and What I've Stopped Using)

One of the incredible benefits of using AI is that AI does not get tired, it does not complain about tedious tasks, and if you can use it effectively, it allows you to free up time, money, and mental effort. Think of an AI agent as a way to cheaply subsidize extra time in your life. There are two major ways that I've used AI, both in my professional career working for other companies, and as my own business contracting with companies. I've been able to build custom AI agents to complete common tasks like taking PDF documents or unstructured data and then turning them into large bodies of structured data that an AI agent can quickly find, saving lots of time in searching for the answer to a known problem. One AI agent that I built regularly found the smoking gun to an environment that was completely down, either preventing a company from seeing customers or preventing a company from being able to sell a product to customers. The AI agent was able to comb through thousands of lines of data in the time that it might take a person an hour or two to do so. These AI agents were also able to watch all the time without taking a break. Something that would be simply too expensive to hire a person to do.

The second way that I've used AI agents is pre-built products like Claude Code. If I have a new problem to solve in say something very complex like cloud engineering, I've been able to use Claude Code to generate scripts for me. They can audit an environment quickly, much faster than I could, and put that output into a specifically structured file, making it very readable for people. I've also learned some handy things like how to give agents code tools, like mermaid diagram renderers, so that they can create accurate and easy to read flow charts. It would be difficult for me to put an exact number on it. But I know that Claude Code has saved me weeks of work.

Here are some of the skills I currently use, roughly in order of how much value they give me:

Cloud Engineer

This agent skill has been a life changer. But this is certainly one of those "please don't try this at home, I'm a professional" scenarios. This tool allows me to connect to my Amazon Web Services Cloud Environment. Now I am a cloud engineer by trade. So I would do a lot of this manually, but I learned that if I connected the AWS command line interface to Claude and gave it a sandbox environment, it was extremely good at deploying just about anything, and it could do it much faster than a person could. It's that simple. There are some plugins now that allow you to do this as a plugin or an MCP. This is a custom skill that I have and one of the advantages of using a custom skill is that you can put your own guardrails in there if you want to. But in my case this was actually something that I built before the plugins were a thing, and you could do it with any cloud provider, Google Cloud, or Azure.

Graphic Design

Allows my agent to do image cropping, photo prep, and create designs and documents with the Typst language. I've made everything from edited photos for a website, to board game cards, to nice looking instructions with this skill. The agent understands how to write in that code-to-text layout language, Typst, and how to render SVGs and edit images because of this skill. Without it, it wouldn't be able to do that very easily, but now any sort of graphic design project I might have, I don't have to go back and try to learn Photoshop again. I just give it the skill and I'm ready to go.

Telegram

Before the advent of plugins with these agents, I had to build this one custom with my coding agent, but I gave it a bot account and API key on Telegram, which allows me to have my agent send me messages for certain things. And I use this skill all the time. If I'm gonna run a long job that may take hours, and I often use my agent on my home computer, I might leave and go somewhere else. And while I'm out, it could send me a report on what happened and what it did. At one point I was even able to rig this up so that I could send it commands and chat with it one on one in Telegram to complete the project. I wouldn't recommend this unless you are pretty capable technically because this will open you yourself up to security vulnerabilities. And there are many plugins now that allow you to more safely authenticate to whichever app you're trying to use for messaging. And many of these agent applications now allow for a multi-device session, so you can start on your computer and pick up on your phone. I find them to be finicky at the moment and not extremely reliable, so I tend to use my own tools instead of the provider tools.

Mermaid Diagram Renderer

This is a skill that allows Claude to write Mermaid diagrams, which is a markup language of sorts that can then be rendered into a chart or diagram. So if I have all of the things I need for a complex diagram or even a picture or instructions, I can just give that to Claude as a prompt or a file, and it can almost instantly convert that into the Mermaid markup language, and then render it as an image file. And because it has the vision capability inherent in the agent, I'll often tell it to visually inspect its output before it gives it back to me and make sure that nothing's overlapping or is hard to follow. The agent often does not do this on its own, and this is another one of those symptoms of the agent being an intern.

Nano Banana

I have an account set up with an API provider that has a connection to Google's Nano Banana image generator. And there's some code that allows my agent to make a call to Nano Banana to create an image. So the agent just has to give a prompt of some kind to that model. So it's an interesting situation where one model is prompting another model. This actually happens pretty regularly in advanced agent engineering. In this case, the agent that I use actually does a stupendous job at being very clear and specific about the type of image I want generated. I use this for doing prototypes or instead of using stock images for a prototype of really anything, but you could imagine a website or a product. I use this Nano Banana skill instead of creating a stock photo or a demo photo for something that may never come to fruition. Of course, this never ends up in the final product of anything. People can almost always spot a fake. This is just a way for me to rapidly prototype certain ideas and get a feel for what they might look like in the real world if I were to pursue beyond this phase.

Job Board / Social Media

There's a certain work and career related social media that I don't often like to check but I do want to keep my finger on the pulse either of opportunities or what people are talking about on the social media site. So I created a skill using Playwright, which is the captive browser tool where the agent can spin up a browser and if I have signed in to an application on that browser, it can save the cookie and access that site later. And it can interact with the site pretty much like a human being would. And I can give it specific things that I want it to check for. So in my example I had it check for the top three posts related to things that I'm interested in, and then had it do some research on each of those to see if what people were talking about was factual and what part it was playing in the news cycle. I could also have it check job opportunities for specific parameters for me. This would be especially useful if you're on the job hunt or looking for new clients, and it would be built into your agent subscription. This is a very useful tool, and you could use it for lots of different things. Just about any website and just about any social media that you wanted to, although beware you might be violating their terms and conditions which might get you banned from the site or your account suspended if they think you are inappropriately using a bot. For myself, ethically, I believe it's wrong to use a bot to post things, and certainly wrong to write my posts for me. This is something that I use to summarize or collate data or analyze the world, not to do creative tasks. That's what people are for. And trust me, other people can tell when you're using AI to write your posts, and it is beyond tacky.

USB Camera

There is a USB webcam that I plugged into my computer, and I asked my agent to see if it could connect to the camera, fire off a photo, and then visually look at the world through that webcam in a still photo. To my surprise, in less than two minutes, it was finished. It connected to the webcam, wrote a small script to fire it, and then was able to read the image visually. This isn't a skill that I use often, but if there's something that I want to show Claude, this is a very quick and simple way for me to do it. I often leave it unplugged unless I'm using it specifically at that time.

What I've Stopped Using

Don't bother with buying templates or so-called prompt libraries. All you have to do is learn the principles and you will be just fine. It's okay to look at other people's examples, of course, but there are people, some very unscrupulous ones out there, who would love to sell you the snake oil of a dictionary of prompts that they use, which is totally unnecessary. The whole point of LLMs is that it is making this speech-to-code or speech-to-computer action conversion very very cheap. Which means those prompts are worth essentially nothing on their own. If someone wants to show you a process for how they do something repeatedly, that's one thing, but don't go buying a small compendium of prompts. The LLMs have gotten so smart that they don't need a lot of special prompting. They just need you to be specific.

---

Want to see how I'd wire up tools for your specific use case? Learn more about Agent Wrangler or book a call.