Intro — set the scene
Three days, one AI assistant, and a dozen completely different outcomes. Most of them were not on my “core skill” list before I started.
This wasn’t about testing the limits of AI’s creativity. I was just surprised at how much I was able to get done in different expertise domains — real-world AI applications — over the past few days when I knew how to drive the tool.
1. Making the Smart Home Actually Smart with AI
My downstairs thermostat was the problem: it had no idea what was happening in my upstairs room. AI helped me:
- Research compatibility between Aqara sensors, Home Assistant, and Voice Monkey.
- Write the automation logic to pass temperature data between devices that weren’t designed to talk to each other.
- Troubleshoot configuration mismatches faster by asking pointed, constraint-rich questions.
Result: Now my downstairs Amazon thermostat responds to my upstairs temperature, using only a small, battery-powered sensor.
2. Handling Files Too Big for an AI Context Window
I had a single file well beyond the 128k context limit of my largest models. Instead of fighting with truncation or manual splitting:
- The AI helped me design a script to automatically chunk, label, and process content.
- We iterated on parsing logic until each chunk could be processed quickly and within the limitations of my GPU memory without spilling over to CPU or system memory.
Result: That script now lives in my personal toolkit, ready for the next time I run into context limits.
3. Rapid AI Prototyping for New Projects
I’d been meaning to build a red-teaming tool, but hadn’t touched the codebase yet. In one session:
- The AI and I outlined the minimum viable features.
- It scaffolded the initial modules, and I adjusted structure and naming to match my style.
- We discussed test cases before writing them, cutting debugging time later.
Result: A working framework I can now expand without wondering if I chose the wrong starting point.
4. AI-Powered Meal Planning and Grocery Optimization
I’ve never done full dietary planning before. With AI:
- I set constraints (budget, flavor profile preferences, variety, healthy, ingredient availability).
- It generated a plan for bulk cooking.
- I had it create the shopping list, and compare prices at different stores for best deals.
- At the store, I got quick clarifications on ingredients I'd never heard of before via mobile.
- We even adjusted batch cooking steps to account for mistakes I made along the way.
Result: The plan worked, the food tasted great, and I've got food prep taken care of for the next month.
5. A Few More Wins
- Installed and configured Conda for a dual-Python setup without a single dependency headache.
- Fact-checked multiple articles for friends, citing sources and cross-checking against credible references.
- Created new workflows for both prompt creation and modular program writing.
The Common Thread
Different domains, different tasks. Same underlying approach for each:
- Set the context. Clear definition of what my goal is, what I already know, what’s missing, and what matters most.
- State constraints clearly. Formats, limits, available tools, and end-state requirements.
- Iterate quickly. Get a first pass, then refine and tweak to move model responses towards the output I needed — a core part of effective AI prompt engineering. Don't be afraid to throw a response or thread away when it feels like the AI isn't quite getting what you're asking.
These tips help make AI practical. And once you’ve seen what that feels like, it’s hard to go back to the old way of chipping away at a problem alone.
Closing
Three days wasn't enough to gain a mastery in each domain I touched, but it was enough time to do some real and meaningful work in each when using the right prompting. The lift was much easier than it should have been, especially when starting from zero — a clear example of how to use AI effectively when you know the right prompting approach.
Want help applying this approach on your team? Explore What We Do, check our Offerings, or book a free consult.