My Approach to AI
I view AI as a tool that can accelerate development, research, and problem-solving, but not as a replacement for human decision-making. While AI can generate code, ideas, and content quickly, I believe it is still my responsibility to understand, review, and validate the output before it becomes part of a project.
My goal is not to remove humans from the process. Instead, I try to use AI where it provides the most value while keeping human oversight at important decision points. The output may come from AI, but responsibility for the final result remains with me.
Why Validation Matters
One of the biggest lessons I have learned while working with AI is that generating output is easy. Determining whether that output is useful, accurate, or aligned with the intended goal is much harder.
Because of this, I treat AI-generated content as a starting point rather than a final answer. The value comes from reviewing, testing, and refining the output until it meets the needs of the project.
Example: Discord AI Chatbot
One example of this came from my Discord AI chatbot project.
The goal of the project was to create personality driven AI characters that users could interact with inside a Discord server. Rather than writing hundreds of predefined responses, I used Ollama to self host models and provided a character description through a personality prompt. The rest of the prompt came directly from the user’s message, allowing the model to generate responses dynamically while following the character traits I defined.
One of the biggest challenges was that different characters required different prompt structures and descriptions to achieve the intended behavior. While the models were generating responses successfully, those responses did not always match the personality I was trying to create. Some characters felt too generic, while others emphasized the wrong traits or failed to capture the behavior I had in mind.
Instead of accepting the responses as correct, I reviewed conversations, adjusted personality traits, refined keywords, and rewrote character descriptions until the responses better matched the intended character. Each personality required its own iteration process, and even small changes to the prompt could have a noticeable impact on how the character behaved.
This reinforced an important lesson for me: AI systems do not automatically produce the desired result simply because they generate an output. Human validation is still necessary to determine whether the output actually meets the goals of the project. The final personalities were the result of repeated testing, refinement, and evaluation rather than a single prompt generating the perfect response.
How I Use AI During Development
I use AI differently depending on the type of work I am doing.
For programming and technical tasks, I always map out the architecture and structural logic first. Once I’ve laid down that foundational system design with pseudocode, I use AI to assist handle the implementation and at quickly debugging so I can fix them. I never treat generated code as automatically correct. I review, test, and modify everything to make sure it actually fits into the broader system I built.
For creative and design work, AI acts purely as an early-stage brainstorming tool. I prefer to handle the layout planning and greyboxing myself to establish the core experience, and then let AI help explore possibilities or prototype ideas within those boundaries. Once a solid direction is locked in, I want the final creative work to come from humans. Human created art and design are something that AI just can’t replicate, so I keep the final execution firmly in human hands.
Personal AI Ethics Criteria
- For AI-generated technical content, I define requirements and constraints before generation, validate outputs through testing, and accept responsibility for the final implementation.
- For visual and creative content, AI is primarily used during brainstorming and early ideation. Final artistic direction and project identity remain human-driven.
- If AI plays a significant role in a feature or project, I spend additional time reviewing and validating the result before considering it complete.
What I Have Learned
Working with AI has reinforced the importance of critical thinking and validation. The most useful AI outputs are not always the fastest or most impressive ones, but the ones that can be understood, verified, and integrated into a project responsibly.
My experience with projects like the Discord chatbot has shown me that successful AI systems are rarely the result of AI operating independently. They are the result of humans defining goals, setting constraints, reviewing outcomes, and continuously improving the system through iteration.
As AI becomes more common in software development, I believe the ability to evaluate and validate AI-generated work will become just as important as the ability to generate it in the first place.