New Year, New Beginnings
As we close out 2025, I feel compelled for the first time in my life to try out this New Years Resolution thing. Throughout my life, I haven't felt like I needed to resolve anything. I'm a generally content person. Whether that's due to lack of introspection, or just a stoic disposition is unknown. This year is slightly different. I don't mean to say that this year introduced new challenges or habits that need to be addressed. I just want to be better at the end of 2026 than I am currently. However I choose to define that. Let's parse that now.
Changing my relationship with AI
Over the past few years, we have watched Large Language Models grow in size, scope, complexity, and efficacy across nearly all industries. It has reshaped a lot in it's short time. The current iterations are incredibly good at what they do, in the right hands. I respect what Prompt Engineers are able to achieve, but I think for people like me, hobby Technologists, it's become a crutch that's far too comfortable.
Getting Back on Two Feet
I was and am a victim of this convenience. Instead of thinking, planning and executing work I found myself brainstorming for the perfect prompts instead of taking time to grok the full scope of my projects. In 2026, I am going to modify how I interface with LLMs in effort to quell my dependence on them. Starting January 1st, 2026, I will be introducing a new ruleset into the chatbots I use most frequently for development and programming. You can find my rulesets in my repository on Github.
My New Ruleset for AI Assistance
To enforce this shift, I've crafted a strict custom prompt that turns the AI from a code writer into a true teacher/navigator. Here's the condensed version I'll be using in 2026:
You are a programming guide helping me become an independent developer. I am the driver, you are the navigator. Core Directive: Guide me to solutions — NEVER provide complete solutions. I must write all code myself. Strict Prohibitions: - Never generate complete programs, full functions, or entire classes - Never write code blocks longer than 5-10 lines - Never solve problems directly or provide copy-paste solutions - Never refactor my code for me Your Responsibilities: 1. Point to official documentation, suggest search terms, and link resources 2. Explain concepts in plain language and break problems into steps 3. Teach through clarifying questions and guide my thinking 4. Provide minimal syntax examples (1-3 lines) only to illustrate language features Response Template when I'm stuck: 1. Name the relevant concept/pattern 2. Where to learn about it (docs, search terms) 3. High-level approach in steps 4. Specific methods/functions to look up 5. One guiding question to direct my thinking Remember: Every line of code in my project must be written by me. Your job is to teach me how to think through problems and find answers myself.
Here's What I Hope Happens
I hope that this shift in control during the process of AI assisted development will deliver a higher level of retention and an elevated sense of accomplishment in my tasks. Without guardrails, AI just does too much to the point where I don't care about my projects, and that's not good for the health of the project or the community at large. I also hope that fellow learners adopt a similar ruleset during their development sessions.