I'm a relatively new programmer with a basic understanding of Python. I first started coding about two years ago, but I took a long break during university, so I have roughly ten months of hands-on experience. I've built a few small Python projects, but I'm still unsure how to plan and structure a larger application.
I'd like to create an application that overlays physical and transition climate risks onto bank balance-sheet data. My first thought was to use Mapbox, but I don't know whether this should be a web application or a desktop application, or what the practical difference is between those options.
I'm feeling overwhelmed by the number of technologies people recommend, including HTML, JavaScript, React, Flask, and Django. I'm not sure what to learn first, so I've made very little progress and feel trapped in tutorial hell.
There's also a lot of discussion about AI agents and the idea that developers should no longer write applications from scratch. Does that mean using an AI coding service, or building an agent myself with tools such as LangChain? How much should I rely on AI while I'm still learning the fundamentals?
4 Answers
The project may simply be too large for your current experience level, so reduce it to the smallest useful experiment. For example, start with a program that loads climate-risk data for one location and displays or summarizes it. Don’t worry about bank data, multiple risk categories, authentication, polished maps, or commercial features yet. If that is still difficult, reduce the scope again.
It’s also worth deciding why you’re building it. If the goal is learning, an existing commercial product doesn’t matter—you can focus on the specific skills you want to practice. If the goal is a business, research what existing tools already offer and identify a genuinely useful gap, such as a different way of quantifying transition risk.
You don’t need to abandon fundamentals because AI tools are popular. An AI can generate code from a prompt, but that is different from understanding, designing, debugging, and maintaining the code yourself. Without that understanding, it’s easy to accept poor architecture or incorrect output without noticing.
Using AI for a project outline, explanations, boilerplate, or a small piece of mundane setup can be useful. While learning, have it help you plan before asking it to implement things, and make sure you can explain and test whatever code you keep.
That makes sense. I’ve mainly been hearing the strongest claims about AI agents from technology executives, so I probably need to treat those claims as advocacy rather than universal advice.
Try building a narrow end-to-end prototype instead of learning every technology in advance. Start with one dataset, one location, and one basic output. This gives you a complete working path and helps you discover what tools you actually need.
For this kind of project, a simple web app may be the easiest starting point because it can present a map and data in a browser, and the development loop is relatively quick. You could begin with Python and a lightweight framework, then add JavaScript or a more advanced frontend only when the basic version requires it. A desktop app is not automatically better; the right choice depends on whether you need browser access, local-only processing, special device integration, or other desktop-specific capabilities.
Keep the feedback loop short. Break work into small tasks, add simple assertions or tests, learn to use a debugger instead of relying only on print statements, and finish imperfect milestones rather than endlessly polishing one part. Understanding core ideas such as functions, data structures, client-server architecture, HTTP, and databases will transfer between languages and frameworks.
I appreciate the concrete approach. I’m going to define a very small first milestone and focus on completing that rather than trying to choose the entire technology stack upfront.
Use a linear list of small, actionable tasks rather than a huge nested plan. Each task should begin with a clear verb, such as “load one risk dataset,” “filter records by region,” or “display the selected value.” Completing these small steps creates a faster feedback cycle and makes the project feel manageable.
You can also ask an AI tool to compare web and desktop approaches, suggest possible libraries, and turn your idea into milestones. At first, ask it to produce a plan and explain the trade-offs instead of immediately generating the whole application.
A good way to stay focused is to define the single most important feature and the reason you want to build the project. Whenever you get distracted by a complicated interface or a new framework, ask whether it is necessary for that core feature. If not, put it aside.