I’ve wanted to write something about AI and how it’s going, it’s taken a while for me to collect my thoughts together. I’ve seen how people use it, why they use it, and I think everyone’s mindset is wrong about it.

Please treat this as an observation, from a hobbyist and someone that works/uses AI in Tech. I invite you to share your opinion or even challenge it, as discussion is where we can evolve.

The Issue Link to heading

You hire a builder and tell them to “sort out the kitchen.” They’re skilled, they crack on, stuff gets done. But if you haven’t told them what you actually want, you might come home to a perfectly tiled floor exactly where your island was supposed to go.

If you didn’t understand what I’m trying to get at, it usually comes down to the way people prompt AI. I get it, sometimes when I type out a lot of information and requirements, it feels like it may not even pay off. The only good thing for me in those cases, I cut it and paste it into Obsidian (or a note-taking app) and use that as my ’to-do list’, or I start to reply to myself. I believe this is essentially the same methodology as the Rubber Duck Debugging, except with AI it talks back to you.

Accessibility Link to heading

AI has managed to lower the barrier of entry into something like Software Development, and I’m okay with that. As someone that had to sift through StackOverflow, Reddit posts, and spent hours of Googling, I completely get why you’d use AI to do it for you. I think in a sense, everyone should be able to express their creative ability, whether that is Image Generation or Code Generation.

The problem, is that everyone has almost been given a tool with zero training or understanding. “Just Ask ChatGPT” TM. People asked simple questions and got impressive results, which leads them to ask more questions about the result, and it flows like a conversation. You no longer read a conversation that two people had on Reddit or StackOverflow, you can now directly ask and discuss it. To me, this has been one of the most enjoyable parts of AI. The time saved on utilising this, versus the searching I mentioned, is so much easier than before.

At the same time, it’s not like the chat window ever tells you to be detailed or give it a well thought out description. Instead you get prompted for things like ‘Ask me a question’ or ‘Want me to find out the weather today?’

The Right Mental Model Link to heading

I use AI primarily as a teacher and a planner. When I present this concept at my workplaces, I advocate for a specific method for coding:

  1. Write a detailed prompt outlining what I want to achieve, how I think it should be achieved, and any specific constraints or concerns.

  2. I use a customised planning agent that returns a strict roadmap, not a wall of code. It looks like this:

    Change 1 - Database Connection for User Login

    1. Action: Create a new secure route to authenticate the user’s credentials against the database.
    2. Reasoning: The frontend is currently passing data, but the backend isn’t verifying it, leaving the app vulnerable.
    3. Outcome: A secure handshake between the client and server, allowing access to the dashboard.
    4. Hint: Focus on the auth_controller file. Think about how the tokens are being handled in the headers.

  3. I investigate and action it myself.

This method gives me a clear idea of what I need to do and helps me understand why and how I am achieving my goal.

Think back to the builder analogy. By generating a plan first, I am handing the builder a precise, architect-approved blueprint. There is zero ambiguity about where the kitchen island is going.

As a bonus for the “vibe coders” out there, this method actually optimizes both cost and output. You use a highly capable (and expensive) model like Claude Opus or GPT-5.5 to generate the flawless architectural plan. Then, you hand that strict plan over to a smaller, cheaper, faster model like Claude Sonnet or DeepSeek V4 Flash to execute the code. You get top-tier quality at a fraction of the API cost, with a much lower chance of the model hallucinating.

My point is, AI is an incredible resource for an on-hand teacher that can help you fight through a mess of thoughts and give you a clear path. How you walk that path is entirely up to you.

The Misconception of Productivity Link to heading

CEO’s love this term, ‘Makes users more productive’ because in their eyes, that equals money. What you’re actually losing, is the critical thinking that your employees have.

A Senior Developer versus a Junior Developer, the Senior will know how to do a simple task and immediately complete it, where a Junior may need to ponder for a while and think about how they’re going to do it. That is based on experience. This is where AI is useful to a Junior, because it can help them figure out where they need to go. It also free’s the Senior’s time for doing something more important, like an architectural decision. Though of course, if I was a Junior, I’d still try run past my plan/code with them to make sure we’re inline with the resolution.

The weakness is where AI is giving you a code block, by clicking copy and pasting it in, it usually means the Human has taken a shortcut to the goal without understanding why or how it works. This is going to have an effect later on in the workplace. Human nature naturally gravitates toward the path of least resistance. We will always take a shortcut if it provides an immediate win, often ignoring the long-term technical debt.

What I’m advocating isn’t slower work. It’s faster learning, which leads to more durable output over time. Just because a product appears quicker than before, you lose the durability of well maintained code in the long run.

Closing Link to heading

So let’s recap on this. People are being given complex tools to accomplish tasks that may otherwise have been out of reach, but they aren’t being taught the correct way of using it. The prompt and context efficiency is key to making the most of it, understanding how and why context rot happens and when to call the conversation quits, and the proper use case for it. It doesn’t help that everything is forcing it in anything they can, and without the proper resources to handle the growth.

Build and spend time writing a well thought and detailed prompt. You will benefit by doing this, it will remain focused and on-track with your own thinking. When you’ve been talking for a while, stop and start a new conversation. Summarise your discussion and hone in on something you want to explore further. If I was talking about replacing a door previously, now I want to know about what hinges to use. Build up your own context, and precisely target something about what you’ve learnt.

AI has no agency, no desires, and no purpose. Without purpose, what is AI actually going to do?

I’ll end it with this prompt: “Make Minecraft 2, using C# and make no mistakes.”

What do you think it’ll put out?