Survival Mode vs Self-Reflection: Two Different Ways of Thinking

Last week, after a brief discussion through messages, I joined a meeting about using AI agents and large language models more efficiently. I entered the discussion with a very specific problem in mind: I wanted to understand how I could reduce token usage while still producing distinctive and intentional UI designs. One frustration I have with AI-assisted development is that generated interfaces can easily become visually generic; after seeing enough of them, you can often recognise the familiar “vibe-coded” aesthetic immediately. That is not what I want for the applications I build. I want each application to have its own visual identity, design language, and character rather than looking like another variation of the same AI-generated interface.

I already understood several ways to reduce unnecessary token usage when working on functionality, backend logic, and repetitive development tasks. What I had not figured out was how to achieve similar efficiency when communicating visual direction to AI. The aesthetic I imagine in my head is not always easy to translate into words, and even a reference image may only represent part of what I actually want. What I was really looking for was a way to compress an aesthetic language into something AI could understand consistently without requiring long prompts and repeated corrections. When I was told that there might be a way to approach this, I became genuinely interested. If someone had discovered a better method, I wanted to learn it and incorporate it into my existing development workflow.

During the meeting, however, I gradually realised that the discussion was focused on something slightly different. The main topic was how to use AI more efficiently by reducing hallucination, preventing agents from going down unnecessary rabbit holes, and structuring their execution more effectively. These are useful subjects, particularly as AI agents become more capable and development workflows become increasingly complex. The discussion also touched on harness engineering and broader ideas about how we should structure our interactions with AI. None of this was irrelevant or unimportant. It simply did not directly answer the particular question that had brought me into the discussion.

That distinction became clearer as the conversation continued. The other perspective seemed more interested in understanding AI systems, their behaviour, and how we should think about working with them. My interest was much narrower and more outcome-oriented: how do I make AI produce the specific result I want, more efficiently? These two interests overlap, but they are not identical. One explores the system itself, while the other treats the system primarily as a tool for achieving a particular outcome. That conversation eventually made me realise that the difference was not really about AI at all. It reflected a much broader difference in how people approach problems.

Through this discussion, I realised that some people naturally spend more time examining the meaning and context surrounding an experience. Instead of immediately asking, “What can I do next?”, they may first ask, “Why did this happen?”, “Why am I responding this way?”, “How have previous experiences shaped this decision?”, or “What does this tell me about myself?” They may explore the assumptions behind a decision before deciding which action to take. Personal history, emotional context, values, identity, and patterns of behaviour can therefore become important parts of understanding the problem. This approach can produce insights that a purely action-oriented mindset may overlook.

I can understand this way of thinking intellectually, but it does not come naturally to me. Long philosophical explanations can sometimes leave me searching for the central point because my mind is already asking what I am supposed to do with the information. If the answer eventually leads to an actionable conclusion, I instinctively want to reach that conclusion sooner. This does not mean reflection has no value, nor does it mean action is always superior. Reflection can help people understand patterns, avoid repeating mistakes, and make decisions that are better aligned with their values. The difference is simply that some people need understanding before action, while others naturally seek enough understanding to enable action.

I sometimes describe my own approach as “survival mode.” I do not mean that in a clinical or psychological sense, but as a personal description of how I have learned to approach difficult situations. When circumstances require decisions, my instinct is to identify what remains within my control and work from there. Spending too much energy on something I cannot change feels inefficient when there are still things I can change. My attention therefore moves toward adaptation, available resources, and the next decision. Over time, that way of thinking has become almost automatic.

My upbringing probably contributed to this mindset. In my family, education was supported through school, but tertiary education required considerably more independence. If I wanted to continue studying, the practical question was not why someone else could not finance it; the useful question was how I could make it possible with the resources available to me. The options might include applying for an education loan, working first and saving enough money to continue later, or deciding not to continue at all. None of those choices is necessarily easy, and identifying options does not magically remove their consequences. But once one option is unavailable, I find it more useful to work with the remaining possibilities than to spend too much energy wishing the unavailable option were different.