How I Use AI Every Day: Systems Over Shortcuts

Most people assume that heavy AI users are using it to cheat or skip work. My use is essentially the opposite.

I am twenty years old, a junior at UNC Charlotte studying data science with an AI minor. I run a web development agency. I am building and maintaining Propify as a full-stack product. I completed a data analyst internship with real deliverables. I am in a fraternity. I play soccer and basketball. I train. I take a full course load.

AI is not how I skip the work. It is how I manage a workload that would not otherwise be manageable.

The Core of the System: TASK.md

Almost every project I build starts with a TASK.md file. Not a vague outline. A specific, structured document that describes the exact thing I want to build. File structure, component behavior, design system, animation specs, deployment steps, and explicit constraints. Written before a single line of code exists.

The reasoning is simple. Claude Code is exceptional at building exactly what you describe. If your description is vague, the output is vague. If your description is precise, the output is precise. That is not a limitation of the model. It is a law of communication that applies equally to working with people. The TASK.md forces me to think through the entire build before starting, which catches problems early and gives the AI a reliable spec to work from.

I use this workflow for every serious project. The notes site you are reading this on was built from a TASK.md. Propify's migration from Streamlit to FastAPI and Next.js was driven by a TASK.md. The Higgins Digital site rebuild had one. It is the single highest-leverage habit in my workflow.

Prompt Engineering Principles I Use

Specificity over generality. "Build a dark, minimal landing page for a construction company in Charlotte, two-column hero, white headline on black, three service cards, contact section, no gradients, no stock images" produces something useful. "Build a landing page for a construction company" produces a template.

Negative constraints. Tell the AI what you do not want. "No purple gradients. No card designs with drop shadows. No Inter font." This eliminates entire categories of generic output and forces more intentional decisions.

Reference-driven prompting. Pointing at real examples gives the model a visual anchor that is more informative than purely descriptive language.

Role-setting in system prompts. For the portfolio assistant at portfolio.davishiggins.com, I wrote a detailed system prompt that gives Claude the full context of my background, projects, and technical skills. That context changes the quality of every response significantly.

Critical Review Is Not Optional

AI-generated code needs a human who understands what it is supposed to do.

I review every significant output before it ships. Not just for bugs. For correctness of approach, appropriateness of architecture, and patterns that technically work but create maintenance problems later. Catching those requires understanding the system well enough to recognize them.

If you do not know enough to review the output critically, you need to learn more before relying on the output. The tool raises the ceiling for people who already have a foundation. It does not build the foundation for you.

The Daily Workflow

Morning: identify what needs to ship. Set context in Claude Code with the relevant TASK.md or project background. Break the work into discrete tasks. Build in chunks, review outputs, iterate. When I am uncertain about an architectural decision, I talk it through before implementing. Not to get the answer, but to pressure-test the thinking.

For research: Perplexity for fast lookups, Claude for synthesis and analysis of longer documents. For documentation: Obsidian with templates I have refined over time to capture decisions and context alongside the work itself.

The model I keep returning to: AI is a power tool. A skilled builder with a power drill does better work faster than a skilled builder with a hand drill. But a builder who does not understand what they are building does worse work with a power drill than with anything else. Skill first. Tools second.