Zhi SangAI has grown from chatbots into agents that can call tools, operate computers and even participate in the physical world through robots. In just a few years, the range of things it can do has expanded considerably.
Enter a project through your own interests
A beginning developer does not have to start by immersing themselves in pages of official documentation. They can enter a project through something that interests them, start building with AI assistance, and approach the documentation through concrete questions.
Using AI does not replace your own thinking. How you deal with a problem reveals how well you can work with it.
Ask the AI to explain the principles, and analyze the complete chain yourself, rather than only describing the symptom. An AI may fix the problem on its first attempt while making a trade-off somewhere else. As the project grows in scope and complexity, failing to understand its internals and execution logic can turn those black boxes into an obstacle to continuing.
For me, an explanation is followed by examining code, logs and actual results: how the calls connect, what the change affects and why it works. That understanding helps me make the next decision.
Discover needs and problems through repeated use
A system is rarely fully considered in its first version. Needs and problems emerge as you keep using it.
When a new requirement appears, I also ask how AI’s reasoning, natural-language understanding and tool use could help make the process more automatic.
For example, tools turn slides into images, and the AI decides whether a contact sheet is readable or a full page is needed. For video notes, it can infer where a picture would help, extract candidate frames, inspect them and choose one.
I want these abilities to participate in the real workflow: tools execute clear, repeated operations, AI makes content-dependent decisions, and I check the result through use.
AI is the engine; it also needs context and tools
Another thought I keep returning to is that AI is an engine. To make it into a usable vehicle, you need to add other things. The context, tools, permissions and execution process around the model are often called its harness.
To bring AI into your work, give it the relevant context and suitable tools, and let it access material close to what you have. Then it can use its own understanding and judgment to help with the task.
I am trying this in several projects:
- A local Agent context system preserves project background, progress and history, then uses a Hub and layered indexes as projects grow.
- Cloud AI working with my local computer uses a tunnel so web AI can read local material, change files, run commands and show results in the conversation.
- Learning automation provides conversion, retrieval, transcription, time lookup and frame tools, with conventions for the AI to inspect material and organize notes.
These are things I am still practicing and exploring. I start from my interests and needs, use AI to begin building, then keep understanding, using and adjusting what I have made.