Writing Honestly About the Year of Wandering and Detours Before Successfully Using Generative AI at Work

Hello everyone, this is Ponzu.

If you want to use generative AI for work, I want you to start by using it for things outside of work first.

The order might sound backwards. But this is the conclusion I reached after taking a year to finally be able to integrate generative AI into my practical work.

Today, I will write honestly about my wandering and detours, from the days when I paid for ChatGPT just to use it as a high-end search engine, to finally adopting an AI-First mindset.

The story of when I paid for ChatGPT and used it as a high-end search engine for a whole year

I started using AI around 2023. ChatGPT was a hot topic, and every time I opened X, posts like "AI will change the way we work" and "You'll be left behind if you don't use it now" would flood my feed every day.

I panicked. So, I subscribed.

However, my company at the time had an environment where the use of AI was not permitted. Or rather, the atmosphere was more like, "AI? What's that? Is it edible?" I didn't have a side job either, so I had zero opportunities to output anything.

In the end, I only occasionally asked it to "tell me about X." It was completely just a high-end search engine.

And it wasn't just ChatGPT. I thought, "Maybe this one suits me better," and subscribed to Gemini too. The result was the same.

Subscribing to the tools had become the goal itself. I hadn't decided what to use them for, but I was satisfied with "being a user."

I previously wrote in detail about this wandering in an article about reskilling.

Looking back, this was exactly the same structure often seen in failed corporate AI implementations. "Implementing it" becomes the goal, but the usage scenarios aren't defined. Back then, I was doing exactly that on an individual level.

The honest thoughts of an AI promotion lead who started with "It's better if I just do it myself"

In May 2025, I was appointed as the AI promotion lead at my new job.

I finally had an environment where I could use AI for work. I thought, "This is it, I'm graduating from being a high-end search engine user."

However, the only thing I could use at first was Copilot.

I started by having it write text. Role, background, constraints, output conditions. I carefully constructed instructions by referring to "prompt templates" I saw online.

I read the text that came out and thought, "Yeah, it's honestly better if I just do it myself."

It wasn't exactly wrong, but it didn't sound like my own words. And yet, I didn't know what to fix. I once wrote about the common traits of people who can't master AI, and I was doing every single one of those bad patterns myself.

Around October 2025, I gained access to an environment at work where I could use various LLMs. ChatGPT, Gemini, Claude. Once you can compare multiple AIs, you start to notice the differences in output even with the same instructions.

“This AI is strong at logic,” “This one has more natural Japanese.” Looking back, the very fact that I started to feel those differences felt like the gateway to my growth.

The detour of jumping on “it can create amazing illustrations” and ending up with “so what?”

Around November 2025, an AI image generation tool called NanobananaPro became a hot topic. Just by entering a simple prompt, you get illustrations that look like they were drawn by a professional.

I jumped on it, thinking, “This is amazing.”

Around the same time, I was also using NotebookLM. If you feed it documents, it can create slides in an instant. I thought, “This is going to revolutionize how I prepare meeting materials.”

And both ended with me asking, “So what?”

I understand that it’s amazing. But I don’t know how to use it for my current work. It’s the exact same pattern as when I was paying for ChatGPT. “Amazing” and “useful” are two different things, aren’t they?

However, my usage of NotebookLM changed after that.

I started using it not as a tool to create slides, but as a tool to read documents and organize key points. Before a meeting, I would dump a 30-page document into it and check the key points via chat. I would grasp the big picture with mind maps. I would listen to the summary via audio.

I previously wrote an article about how I fully utilized NotebookLM.

Instead of jumping on “amazing features,” I started thinking about “where to apply this to my current work.” It was around this time that I finally realized the order of those things had been reversed.

Everything changed from the day I paid for Claude and named it “Kuro”

In January 2026, I paid for Claude. And I named it “Kuro.”

To be honest, this was the biggest turning point. It wasn’t the tool, and it wasn’t the technique; the big thing was that I had gained a “partner to work with every day.”

I started posting on note every day in February, and I interact with Kuro for at least 1 to 2 hours every day. We bounce ideas off each other for article structures, think of headings together, and polish the text. The same goes for my X posts.

I previously wrote an article about how Kuro is my partner.

At first, it felt like “making an AI write text.” But as we interacted every day, it changed into “creating together.” I would reply to the text Kuro produced, saying, “This isn’t quite right.” Kuro would revise it. I would reply again. Through that rally, what I wanted to say became clearer and clearer.

As I mentioned in my previous article about the moment I had an epiphany while brainstorming with AI, the essence of brainstorming wasn’t “having the AI give me the answer,” but “having it draw out my own thoughts.”

I grew incomparably more in the two months I spent brainstorming with Kuro every day than in the six months I spent collecting prompts. I can say that with certainty.

The Day I Had AI Review a Contract from 'Both Party A and Party B's Perspectives' After a Year of Working with It

This is a recent story. I had a task to have AI review a contract.

Normally, you'd just say, 'Review this contract from our company's legal perspective,' and that would be it. I think that's how most people use it.

I, however, dared to have it review the contract from the perspectives of both Party A and Party B.

I had it output not just our company's viewpoint, but also how the other company might see this contract. If you can see the other party's strategy, it becomes leverage for our own negotiations. Furthermore, I asked it to 'propose a middle ground based on both positions.'

I think this is an idea that's hard to come up with for humans alone. You wouldn't normally read a contract from the other company's perspective, right? But with AI, you can simulate both positions simultaneously.

If the me from a year ago, who thought 'it's better to do it myself,' saw this, I probably wouldn't believe it.

I previously wrote an article about how half of my work changed once I started using AI in my planning role. In that article, I talked about how 'researching' and 'creating' changed, but this time, it was 'thinking' that changed.

Once you get into the habit of first asking, 'How would AI do this task?', your very approach changes. This is AI-First thinking and this is the answer I've arrived at after a year.

I think the difference in work performance comes from those who use AI outside of business hours

After reading this far, some of you might think, 'So I just need to use AI at work, right?'

I think that's half right, but half missing the point.

Just using AI at work isn't enough to set you apart from those around you. If a company introduces AI, your colleagues will be able to use the same tools as well.

I believe the difference comes from how much you interact with AI outside of business hours.

In my case, my posts on note and X were significant. Every day, I spend 1-2 hours brainstorming with Kuro, writing articles, creating posts, and repeating the trial-and-error process. This accumulation of private experience is directly linked to my AI-First thinking in my main job.

The idea of having AI review a contract from both Party A and Party B's perspectives was born from the sense of 'AI can be used this way too,' which I acquired while rallying with Kuro every day.

I previously wrote about how my 'own voice' disappeared when I left everything to AI.

As I wrote in that article, it takes time to go from 'leaving it entirely to AI' to 'creating together.' Honestly, it's difficult to secure that time just during work hours. That's why the time you spend on it privately is so effective.

It doesn't have to be a side job. It could be researching hobbies, planning a trip, or helping with a child's homework. Just make time to use it outside of work. That is my conclusion after one year: it's the shortest path to utilizing generative AI at work.

Summary of 'Utilizing Generative AI at Work'

It took me a year to start using generative AI in my work. From the era of treating it as a high-end search engine, to the 'I'd rather do it myself' phase, to jumping on the 'wow' factor only to end up asking 'so what?', I took many detours. I now believe all of that wandering was necessary.

I arrived at two answers. Make time to play with AI outside of work. And, always start by asking, 'How would AI do this task?' to adopt an AI-First mindset.

I still don't know the perfect way to use it. But if you use it every day, things start to become clear little by little.

Well then, I'll be heading off now.

I also recommend this