Basic AI Tool Guidance

Notebooks (e.g., Jupyter / Colab)
- Use for: quick experiments, data exploration, and early prototypes.
- Don’t use for: producing finalized products.
- While you can prototype a chatbot in a notebook (and even spin up a quick Gradio/Streamlit UI), you don’t usually deliver a final product as a notebook in the real world.
IDEs (e.g., VS Code / PyCharm)
- Use for: building final products and real applications (with folders, files, tests, Git, packaging, deployment, UIs, and users). This is the ideal kind of tool in which to build a chatbot or other finalized AI product.
- Optional: Add a coding agent (e.g., Codex) in your IDE; it can read/modify code. IDEs and coding agents are professional-grade helper tools, not a final product themselves.
Commercial apps/websites (E.g., ChatGPT / Co-pilot / Gemini / NotebookLM)
- Use for: exploring, drafting, brainstorming, coding help.
- Don’t use for: final products. For final products, do use an IDE to build your own UI and backend code for calling APIs and other AI tools.
- The ChatGPT website is its own UI. Don’t try to “wrap” that UI for a product–that’s silly.
Coding agents (e.g., Codex)
- Use for: generating code, fixing bugs, writing tests, etc.
- Don’t use for: hosting or delivering a user-facing final product.
- These help produce code for your app/product; they are not the app you deliver to end users. Don’t try to wrap coding agent UIs for your products–that’s silly.
Language model APIs (e.g., OpenAI API)
- Use for: your app’s backend data processing—chat generation, retrieval, text-to-code, etc. You build your final product using APIs.
- Language model APIs also mean the difficult computation occurs on someone else’s machine.
- Language model APIs are a general low-level tool for interacting with AI models within your own custom-built applications or products.
Websites, open-source models + tooling (e.g., Hugging Face & transformers or langchain packages)
- Use for: loading and running open source models; your app’s backend data processing—chat generation, retrieval, text-to-code.
- Open source models tend to run on your own machine (you will need better equipment, more memory, GPUs, etc.)
- You can build a final product around/with Hugging Face models with or without commerical language model APIs.
- Don’t use for: “wrapping” the Hugging Face website as your own product–that’s silly.
Agentic tools (e.g., Zapier, n8n)
- Use for: building automated workflows for your own or small group use.
- Don’t use for: building a final product to sell to external users. Like ChatGPT and/or Codex, these are already finalized commerical products with their own UIs.
- Mostly focused on software engineering automation, but can also perform some office tasks.
Connectors (e.g., MCP, ChatGPT Apps)
- Use the model context protocol (MCP) to connect a language model API to various tools (databases, APIs, files, search, internal systems) in a programmatic way; MCP is typically used on the backend of apps or products.
- Use specific connecters (e.g., ChatGPT Apps) to let ChatGPT or other commerical apps connect to files, repos, messages, etc.; This enhances ChatGPT’s abilities, and is not for building apps or products you surface to others.

Usage Rules of Thumb:

To learn and research for yourself: Use chatbots (e.g., ChatGPT, NotebookLM).
To build a product for others to use: Prototype/test/understand coding logic, API calls, and model usage in notebooks → Build a finalized chatbot or other AI app in an IDE, while letting coding agents help write code and build project infrastructure.
To build an automated workflow for yourself:
- Simple: Use Zapier, n8n, etc., or ask a chatbot–you may get lucky.
- Advanced: Use a coding agent to write the software to complete the workflow.
Never input sensitive data.
Don’t draft, don’t code, prompt!

Disclosure: I worked with AI to build these pointers.

Licensing

Text and copy are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0). You are free to share and adapt this material for noncommercial purposes, provided that you give appropriate credit, link to the license, indicate if changes were made, and distribute any contributions under the same license.