TLDR: I share a real example from today when AI cut down 10 hours of work. I discuss some best practices of using AI at work.
My name is Chad and I’m 2+ years into OE, which has let me double my pay since business school. Living remote in Bangkok has given me time for side projects while keeping costs down. I want to teach you how to do the same.
The Data Job That Came Up Today
I am literally writing this as I eat lunch. At J2 today I had to look over transactions from a key client partner. The Excel file was packed with thousands of rows, the kind of file that bogs down my crappy work laptop. I guess you can say this type of work is pretty typical for both my J1 and J2 that are in the consulting industry.
Since I was the only one assigned to the task both now and historically, I could have taken longer, maybe told my manager it needed extra days (see my article on information edges). But I got through it in about 3 hours with help from AI. From what I’ve seen, my J2 firm doesn't mind AI use—I've seen managers do it— and I feel that enforcement is a grey area on purpose. This is helpful for OE, as it means utilizing my full OE toolkit.
I pay $750 each month for ChatGPT Pro ($200), Grok Heavy ($300), and Gemini Ultra ($250). An absurd amount of money for subscriptions that’s for sure, costing about the same as my Bangkok rent. But it gives back value. Staying at the cutting edge of AI that is available to general consumers make OE possible, and ChatGPT sends me weekly notes on the most up-to-date advances associated with AI use in white collar industries.
I think one practice people get wrong is trying to chat back and forth with AI on hard jobs. Instead, I go for one-shot prompts and I recommend you try doing the same, especially for data-manipulation or difficult tasks. You put all the details in at the start, and let it run.
My prompt today to ChatGPT 5.2 on Pro (extending thinking) was close to 600 words, with the 9MB file containing thousands of rows. It took 30 minutes to write, but then AI ran for an hour, and I could do other things (like write this article 🙂 ).
I know this guy named Tom who's also in OE. He has a J1 in project management and a J2 doing some light analytics.
A month or so back, he had to clean up a sales report with mixed-up data from different sources. Tom started by giving AI a basic file and then chatting to fix issues step by step. But after a few rounds, the AI lost track of early details, like specific date ranges he mentioned first.
The output had wrong sums and missed some outliers. When he sent it to his J2 team, they spotted the errors right away, and he had to redo it all by hand.
It made him look sloppy, and he worried about his optics.
After we talked, Tom tried the one-shot way: He wrote out everything upfront, including context on the data quirks. The result came back solid, and he just had to tweak a few spots. It's a small switch, but it shows how iterative chatting can backfire on sensitive work.
Building Habits Around AI Use
I guess this nugget of advice doesn’t apply for just OE but could be used in your everyday lives, irrespective of the context. For complex AI tasks like my Excel review today, aim to one-shot it by thoroughly preparing a prompt
My real prompt from today. Notice the context I provided, my instructions, and warnings of outlier outputs.
Give AI the information that one would need if if you they were to manually complete the task: what the input is, what the output needs to be, what to do with the input, and any special cases like name errors or odd values. Make sure you're on a high-power AI model for the heavy lifting.
During the hour ChatGPT was working away, I planned this article, ate lunch, threw out the trash, and went to the 7-11 at my condo in Bangkok.
Without this, the job would have eaten up at least two full days. I spent the bulk of my time getting data ready and checking with the team on things like scope and outliers. The wait was 1 hour, and then 30 minutes to clean up. In fact, I didn't send it out right away—waited until the next day, which gave me time for J1 stuff and writing this.
Practices That Have Made a Difference
Here are some things I've learned that can help.
Get the full context first. Make sure you understand the task completely. Write it down so you can pass it to AI. I use my phone to record meetings (helpful tool is otter.ai), which is useful when I'm in one but thinking about another job. Your understanding of the context is what makes you stand out—like knowing which files are key or how parts connect. In my experience it's often better to handle things yourself with AI help rather than trying to pass it off due to the contextual barriers to offloading your tasks completely.
Know what AI struggles with. It improves over time. Older versions couldn't deal with big files well. Choose the right tool and give it a test run. (at the time of this writing, I used ChatGPT 5.2, which has improved Excel capabilities)
Give clear directions to AI. List out inputs, what you want as output, and the background. For Excel work, tell it to keep steps simple, maybe use a extra sheet for testing.
Always check the output. Make it look like regular work. AI can make file names or formatting too neat. I go in and simplify things, like formulas or borders.
There was this time a few weeks ago when I was moving between spots in Thailand, with not great internet in some places.
My J1 needed a quick market trends for an upcoming meeting. So I prepped a full one-shot prompt in MS Word while I was in transit, exactly when I did not have signal. I ran it during a stop where signal was okay, and it finished by the time I arrived at my destination.
Thinking About What This Means for OE
OE requires one to do smart work, not avoid work. Look for jobs where you work alone and rules on tools are loose. With tools changing fast, behind ahead of the curve is also important, as firms inevitably catch up to these advancements (and white collar labor requirements decrease).
In the end, working smarter frees up time you can turn into higher comps and better QOL. If you try one small change, like testing a one-shot prompt on an easy task, it could start adding up for you.