I spent over $750 on AI last month, and it still made me look like an idiot in front of my manager.

TLDR: I got lazy on a new J2 project and let an AI build a slide deck without enough context. The result was a sloppy, amateurish failure that got called out.

I've been running OE for over three years now. Ivy League undergrad, M7 MBA, J1 at a brand-name consulting firm, J2 at a boutique shop. I run both from my condo in Bangkok and I've built my entire workflow around hyper-efficiency and AI tools. But last week, that system hit a wall. And it was entirely my fault.

Jumping onto a new project mid-stream is one of the worst things that can happen to your OE setup.

You're dropped into a moving train. There are months of context you don't have, established norms you haven't learned, and a new set of political landmines. This was my situation last week at J2. A big, multi-year client. A project already in full swing. I was a new guy trying to get up to speed while keeping my J1 workload from spilling over.

I usually keep my combined hours under 25 a week across 2 fulltime jobs. This past week, it was easily over 40. I had to skip my Thursday Storytime post last week because of it (sorry about that, by the way). It's during weeks like this that OErs might get caught up making mistakes (as I did ☹ ).

For some context on the project itself: it's a classic consulting gig. Two main workstreams. First, the analysis: taking a mountain of client data from Excel docs and doing simple math, essentially. Second, the stakeholder management: presenting updates and keeping the client happy.

The problem was that already months of methodology and norms were baked into the team's brain but not mine.

A task landed on my plate: build a summary slide for a client update. I thought this was the typical consulting BS busywork, but fair enough.

I was juggling calls between J1 and J2, so I did what I always do. I took the meeting transcript, threw it at ChatGPT (or my tool of choice at the time), and told it to crank out the slide. Obviously, I checked the output. I waited a couple hours before shipping it (a classic underpromise move) and sent it off.

That evening, I saw the team's final presentation and my contribution wasn’t included. My manager had completely rebuilt it from scratch.

He sent me a few lines of feedback that were polite but firm. The colors were a shade off the official client template. The text wasn't calibrated to the project's specific jargon. Worse, ‘I’ used a collection of PowerPoint shapes cobbled together to look like a table, instead of using a native table object.

My gut reaction was to be annoyed at my manager for being so anal about a single slide (in fact, he didn’t mention anything about making it part of a larger file).

But that's the wrong takeaway. And I think it's the wrong takeaway for a lot of people who are starting to lean on AI.

A Gallup study from 2025 found that 66% of remote workers use AI tools. That's up from 28% in 2023. But here's the thing: I believe there's a gap between using AI and using it well. Most people are in the casual-use zone, treating it like a smarter search engine. They think the work is the deliverable. They're wrong.

The real work is building the scaffolding around the AI.

To be completely honest, I would have made the exact same mistake if I'd built the slide manually, outside of AI. The issue wasn't the tool. The issue was that I was dropped into a project with three months of established norms and I hadn't yet built the systems to account for them. I hadn't done the upfront work of understanding the exact specifications my team expected. That's an exogenous shock to my OE setup.

So I didn't just fix the slide. I went through our team's shared files and created my own personal slide-building specification document. Matching the same anal energy of my manager, I documented every single detail: the exact HEX codes for the primary and secondary color palettes, the font names and sizes for titles and body text, the safe-area margins, the border weights for tables, the required styling for header rows and body cells etc. etc. etc. I turned my manager's nitpicks into a machine-readable instruction set.

This document is the leverage. It's the asset in your OE toolkit. It's what turns your intern of a ChatGPT into a more agentic experienced employee while you do other things.

A similar task came up a couple days later.

This time, I fed the AI the raw data. The ChatGPT ‘folder’ already had my specification document. The result was an absolutely perfect slide. Native PowerPoint table, correct colors, pixel-perfect alignment, etc. etc. It took the AI about 30 minutes working by itself using one prompt. I waited two hours before submitting it.

My manager thought I had been manually tweaking shapes. He didn't need to know I was already back to working on my other J, writing up this post for you, or my SaaS side projects. This is the asymmetric information we live for.

(As a side note, the screenshot I attached is from ChatGPT, but I personally recommend more agentic systems over a simple chatbot for heavy-lift tasks. That said, ChatGPT 5.2's recent update has been surprisingly good for white-collar tasks like slide building and data analysis. I sometimes stick to a ChatGPT project because it holds the full context of everything I'm doing.)

I think the broader lesson here is one that a lot of OE skeptics and AI deniers get wrong.

They see a failure like mine and say, "See? AI can't replace humans." That's bullshit. The failure wasn't the AI.

The failure was me, the operator, not giving it enough context. AI is an amplification of how you think about your work. Give it garbage instructions, get garbage output. Give it a detailed, specific manual, and it'll produce work that's better than what most humans would do in ten times the amount of time.

This is about building the leverage you need to chase that bigger bank account and the freer life we all want deep down. It's not about being lazy. It's about being the architect of a system that makes your output indistinguishable from someone grinding 60 hours a week. 🌴

Stop thinking of AI as a plug-and-play solution. Think of it as a new hire. A very fast, very capable, but very dumb intern. It needs to be trained. It needs a detailed manual. It needs a system.

Your job, as an OE professional first, is to be the architect of that system. The time you invest upfront in creating detailed instructions, templates, and specifications will pay you back a hundredfold. And that, my friend, is the real reason your boss doesn't notice you're doing two jobs.

I go much deeper into the AI toolkit I use daily in another post. Check out my article on my $750 AI bill and how I use it to stay ahead.

How would you rate this article? Click 1 (needs work) to 10 (loved it)!