How I Actually Use AI (Not How Social Media Talks About It)
This is an operator’s account, not advice.
There are plenty of stories about how AI “changed everything,” enabled passive income, or replaced entire jobs overnight. This is not one of those stories.
AI didn’t replace my thinking.
It didn’t magically create revenue.
And it didn’t remove the need for judgment.
What it did—gradually and unevenly—was remove friction from work I was already doing.
This blog exists to document that process honestly.
Before AI: High Effort, Low Leverage
Before AI entered my workflow in any meaningful way, most of my work looked the same every day:
- Document-heavy inputs
- Repetitive analysis
- Manual extraction of structured data from unstructured sources
- Constant context switching
Automation was appealing, but brittle. Scripts broke. Edge cases multiplied. Every shortcut created downstream cleanup.
The problem wasn’t a lack of tools. It was a lack of leverage.
My First Real Use of AI Wasn’t Automation
Early on, I didn’t use AI to automate workflows. I used it to think.
I treated it less like a generator and more like a reasoning partner:
- Talking through system designs
- Stress-testing assumptions
- Exploring tradeoffs before writing code
- Replacing long search sessions with focused dialogue
This mattered more than speed. It reduced decision fatigue and made iteration cheaper.
Automation came later.
Letting AI Work Against Me
Letting AI “run free” wasn’t the mistake. Exploration was necessary. The mistake was assuming the problem was control, rather than alignment.
I tried to bend commercial, paid LLM tools into workflows they weren’t designed to support. The limitations surfaced quickly:
- Broken or inconsistent graphics
- Misspelled words and subtle errors
- Visually flawed outputs that required manual repair
- One-off results that couldn’t be reused or scaled
Even when outputs looked acceptable, they weren’t composable. Each run had to be recreated from scratch.
At that point, the work wasn’t with AI—it was cleanup because of AI.
Prompts as a Forcing Function
Trying to make this usable pushed me into an unexpected place.
I started designing prompts not to “get better answers,” but to escape the box the model itself was creating. Prompts became a way to counteract limitations rather than amplify strengths.
That shift forced me to:
- Think more deliberately about structure
- Anticipate failure modes
- Separate intent from what the model could reliably produce
The insight wasn’t “prompt better.”
It was that the model needed boundaries—and those boundaries couldn’t live only in language.
Structure First, AI Second
The turning point came when I stopped asking, “How do I get better output?” and started asking, “What structure does this problem actually need?”
That led to:
- Explicit schemas
- Field maps instead of free text
- Directional extraction rules
- Clean outputs stored independently of their messy sources
Once AI was forced to operate inside structure, it became useful.
Not creative.
Not magical.
Useful.
AI as a Component, Not the Product
Today, AI sits inside my systems the same way a database or parser does.
It has a role.
It has defined inputs and outputs.
It is monitored.
And it is replaceable.
Most importantly, it is never the final authority.
Human review isn’t an afterthought—it’s designed into the pipeline. Every system assumes AI will be wrong sometimes and plans accordingly.
That design choice matters more than model selection.
Why This Blog Exists
This blog isn’t about prompts, hacks, or chasing trends.
It’s about:
- Real workflows
- Real constraints
- What worked, what failed, and why
- How AI fits into systems that already need to function without it
Most importantly, it’s about using AI deliberately—to do more with the same effort, without giving up judgment.
Closing
Even this site is part of the experiment.
It’s built with a human-in-the-loop publishing workflow. AI assists, edits, and structures—but nothing publishes without review. That process will evolve, and I’ll document those changes as they happen.
If you’re looking for instant results, this probably isn’t for you.
If you’re interested in how AI actually fits into real work, you’re in the right place.