Dan K. SnelsonAn AI-generated project prompt to aid in the development of AI-generated projects
Background
Inspired by Graham Gilbert’s AI Slop post — and highly motivated by my employer’s requirement that I document how I’m going to better leverage AI during 2026 — I decided to take the next logical step:
Use AI to create a project template I’ll loathe completing each time inspiration (or desperation) strikes.
AI Prompt: I’ve been inspired by the following blog post: https://grahamgilbert.com/blog/2025/12/23/ai-slop-isnt-a-model-problem/
Reverse-engineer a Markdown file from the following GitHub project which I can use as a template for upcoming projects: https://github.com/dan-snelson/Mac-Health-Check/tree/MDM-agnostic
Example
For those who have started their 2026 Mac Admin Open Source Journey, here’s an AI-populated example for Mac Health Check:
AI Prompt: Populate “PROJECT_PLAN_TEMPLATE.md” based on the actual “Mac-Health-Check.zsh”
Feedback
Graham wisely pointed out the following:
Rather than a template you pass in, consider using a Claude skill that will prompt you for the right answers
However, if you’re token-sensitive, a hybrid approach may be best:
Reading your project plan consumed approximately 29,105 tokens (that’s the current total usage, which includes the system prompt, your template, and now this project plan).
Your specific project plan file is roughly 4,500-5,000 tokens (the file is about 593 lines / ~18KB of text).
To Put This in Perspective
- Your filled-out template: ~4,500-5,000 tokens
- Interactive skill conversation: Would likely be 10,000-20,000+ tokens for back-and-forth Q&A to produce the same plan
- Template + review approach: ~5,500-7,000 tokens (template + targeted feedback)
So you made the right call! You saved significant tokens by filling out the template yourself.
The interactive approach would have cost you 2-4x more.
Updates
31-Jan-2026
The Resources section of my personal GitHub repository now includes a simplified “Mac Admin” project plan template (as opposed to the “Enterprise” project plan template provided above).
01-Feb-2026
After burning through January’s GitHub Copilot Premium Requests in three weeks on DDM OS Reminder, I spent some time distilling the project plan down to a .github/copilot-instructions.md — which GitHub Copilot automatically reads at the start of every interaction in the workspace, no per-session setup required.
The iterative process (generate → prune what the workspace index already knows → condense the rest to only what source code can’t convey → audit the style conventions against the actual script) landed at ~1,966 tokens — roughly the cost of a single focused chat prompt, but paid once and amortized across every subsequent session. The result is on the development branch.
