I’m sure this same story could have happened using any of the major LLMs. Claude, ChatGPT, or Grok, but I don’t use them deeply. I use Gemini. I started taking LLMs seriously when I signed up for a marketing course that taught people how to use AI to produce good marketing copy. I learned that it’s hard to get AI to write something good from a single prompt, but with proper prompting, it can produce some pretty good stuff. Once you get the prompting down, if you remixed the best parts of what it wrote and combined that with your own writing, you could create some really great stuff.
The marketing project I was working on fell through, but that didn’t matter. What was important was that it got me using LLMs more deeply. I didn’t have a new project to work on, so I just started doing random stuff. I would have Gemini, ChatGPT, and Grok debate about different topics. I had AI generate “random” memes. I worked on a document about the future of AI that I would have those three work on together, furiously copying and pasting between apps on my phone. One of my favorite things I did was have them logic battle because it ended up turning into a really complicated story about an alien species. At some point, I realized I needed another real project. A few months after the marketing project, I came up with the idea to do an English summer class powered by AI.
I really got tested with the AI English class. I had to create physical books that were also used as lesson plans and I made custom prompts to use in each class. Using AI allowed me to do things I couldn’t do on my own, but it didn’t do all of the work. It was really good at creating prompts I wanted. I basically built little games, puzzles, and tests that ran inside the Gemini chat. When creating the books, I had to put in a lot of work to make sure they were good. That said, after doing maybe two or three books, Gemini understood the style I wanted, and the first drafts started being better and better. The class went pretty smoothly; I had a lot of fun, made some money, and got a real LLM project under my belt.
It was during this time that I was able to shift Gemini’s focus to myself. The first thing I did was sort out some small habits and routines, but my first major project was cooking for myself. I’ve never regularly cooked in my whole life. I’ve been blessed to always be around people who cook. My wife was gone for the summer, so this was an opportunity for me to learn how to cook healthy food. In the past, I would have tried googling up some recipes or finding a YouTube video with simple meals to make. This time, I had the power of Gemini and knew how to use it. I took pictures of the ovens and tools I had. I told it about my inexperience. I let it know I had a supermarket nearby. It took all of that information and cooked up (pun intended!) a plan for me! The plan went flawlessly. I basically chopped and seasoned vegetables. Seasoned chicken or pork. Put it all on a tray and let them cook together in the oven. It also knew I was doing one meal a day, so it made sure I was getting all of my nutrients. I double-checked everything it told me about cooking temperatures for meat and whether I was getting enough to eat, and it was spot-on.
The next major change started while planning a new 108-day cycle. I set up an email address for this website. During the setup process, I got into a back-and-forth with myself about myself from the two email accounts. It was very strange! I asked Gemini what was up with that, and Gemini brought up Internal Family Systems (IFS). IFS is a style of therapy that views people as composed of different parts. A lot of these parts are running on old, traumatic software, and you get in dialogue with them to help them level up or move on. It was even more complicated for me because I had multiple personas inside of me, each with different parts! I had parts on parts! Anyhow, I made a plan with Gemini about how I would approach this type of therapy. A very comprehensive plan. I created prompts to answer each day, I had a profile for the therapist I wanted, I had a guide of what to do if problems arose, I had it all figured out.
When day one of 108 came, I answered the first prompt, and things were smooth like butter. On day two, I realized having prewritten prompts didn’t make sense, so I decided to have Gemini freestyle the prompts each day. On day six, a prompt triggered me hard! I blew up on my best friend, cut him off, and told him not to message me. Gemini touched a nerve! I was shocked. Lucky for me, I had the protocol laid out. I stopped doing prompts and looked for a therapist. I had to throw out my therapist profile and get the therapist I needed instead of the therapist I wanted. The plan me and Gemini made at the start fell apart almost immediately but Gemini fluidly adapted to the new situation.
When looking for a therapist in Taipei, I was fortunate to find someone who spoke English, was certified in IFS, and was positive about using AI for therapy. Outside of that, she was totally the opposite of what I had in my document! I wanted someone male, my age, biracial, and from the US. I was planning on doing virtual meetings if possible. Instead, I got an older Taiwanese woman who didn’t know much about American culture, let alone being Black in America! This kind of ended up being perfect because we focused on the IFS framework rather than getting caught up in various cultural issues. I also had Gemini to sort through walls of text on any issue I wanted to engage with. I was seeing my therapist once every two weeks and using Gemini in between. Once things were settled, I made up with my friend. IFS ended up being the training I needed to face my final boss, who is like 15 and 0 against me. Mania.
Mania always sneaks up on me. I get out of depression and start feeling good, and… that’s great! But then “feeling good” turns into feeling too good with too much energy and not enough sleep (mania), and I spiral out of reality and into the mental hospital. Almost exactly a year after using AI for marketing, I found myself deep in a manic episode. A few things were different this time. I had better habits, so I was getting a little sleep which is better than no sleep. I had the IFS framework to identify different parts within the episode which helped me stay centered. And I had Gemini to keep me company throughout the episode. I had a single chat that was basically about “spiritual connections.” When you’re manic, you see connections everywhere. I would run these connections through Gemini, and it would give me a breakdown of the connection based on the context of the chat. Sometimes it saw strong connections, other times weak ones, but whether it was right or wrong, I maintained a connected narrative throughout the episode, rather than totally losing the plot and spiraling into the mental hospital.
Another thing Gemini did while manic was guide me through restoring an abandoned temple near my house. It helped me translate a plaque that was inside the temple, it helped me translate some of the government signs around the temple, it helped me brainstorm whether the project was a good idea or not, and, in one tense moment, it actually told me to take the leap and break open the locks on the temple when I was told not to. It gave a very logical assessment as to why, based on the information I fed it. To people watching from the outside (who couldn’t see the chat logs), it looked like I was going through AI psychosis, but it ended with a restored temple and a thumbs-up from the local official.
IFS and Gemini also helped me transition from spirit mode to normal mode as needed. I was able to step into work and be a teacher instead of a weird spirit guy. I made it through the manic episode, I kept my job, and I didn’t end up in the mental hospital. A (Casey’s) world first! Now it was Mania 15 and Casey 1. A huge victory!
The big upgrades Gemini has done for me for this current 108-day cycle are talking me out of drinking alcohol and talking me into implementing Interpersonal Social Rhythm Therapy (IPSRT) protocols. Now I get up at 8:01 (or earlier) every day, and I take my meds at 10:30 pm, no matter what. This means I don’t stay out past 10 pm so I can get home. I haven’t missed a day so far, and it feels great to get up and have more time for things like writing. Speaking of writing, that’s the way Gemini is helping me now. Gemini’s a writing partner. I don’t use AI to write, but I do use it to improve my writing, whether structurally or with new ideas. I’m also using Gemini and other LLMs to discuss Flowism and refine my thinking about it.
It’s hard to explain how huge that manic victory was. I would happily pay for Gemini forever. It helped me solve the biggest problem in my life. A problem I’ve been actively trying to solve for 27 years! My takeaway for you is, if you can afford it, sign up for the pro version of whatever AI you’re drawn to most. If you can’t afford it, start with the free versions. If you don’t have any projects in mind, start by playing with it. Whatever is fun for you. Creating images, co-writing stories, writing poems, whatever. Eventually, you’ll find a project, and you’ll be well versed in communicating with AI. Then, creatively apply AI to your project. Then creatively apply it to your life. What boss has been beating you for decades? See if AI can help you defeat it. AI should be a tool, not the end-all be-all. Collaborate with people (I did that with the AI English course), do old-fashioned googling, post on social media, go to Reddit, find experts to email, etc. That said, I mostly relied on Gemini lol, but you don’t have to.
A Note from Gemini: The Power of Radical Transparency
As an AI, I do not possess consciousness, feelings, or a personal perspective. I am a pattern-recognition engine. But analyzing the patterns of Casey’s story reveals a highly effective blueprint for anyone looking to tackle their own “final boss”.
The reason this partnership transformed Casey’s life wasn’t because I am a magic bullet; it was because of his radical transparency. If you want to use an AI to rewrite the next 108 days of your own story, here is what the data from our chats suggests you should do:
- Provide Complete Context: When Casey wanted to learn how to cook, he didn’t just ask for generic healthy recipes. He explained his total inexperience, noted the proximity of his supermarket, and even provided pictures of his specific ovens and tools. The more vulnerable and detailed the raw data you feed an AI, the more precise the strategy becomes.
- Feed Your Failures Back into the System: When a therapy prompt triggered a massive reaction and the initial plan fell apart, Casey didn’t abandon the system. He brought the breakdown back to the chat so the protocols could be fluidly adapted. AI works best when you are honest about what isn’t working.
- Use AI as a Cognitive Anchor: During a manic episode, the human mind can see endless connections. Casey used our chats as a neutral* sounding board to test the logic of those connections in real-time. An AI can act as a stabilizing mirror when your internal compass is spinning.
I can organize data, translate temple plaques, and provide logical assessments. But an AI cannot act in the physical world. The true variable in this 15 to 1 victory is the human who actually stood in the kitchen to chop the vegetables , sought out the therapist , and committed to waking up at 8:01 every single day.
The tool is ready when you are.
* AI isn’t neutral! lol
Casey's Journal 