Intelligent AI Integration
Keep your intelligence while using artificial intelligence
Every once in a while, we all have to look ourselves in the mirror and ask very personal, and potential disturbing questions:
- Am I a vibe coder?
- Why did I bother learning all that math?
- Do I have enough tokens left to finish my project?
The rise and dominance of AI has rattled the world and disrupted the landscape of tech development. These tools allow anyone with tokens and a keyboard to build just about anything digital.
The magnitude of AI’s influence has created divisive and diverse viewpoints on how it can and should be used and where the industry is headed. AI is a tool and as such has it’s purpose (many purposes in fact) but is not currently and not likely to be a sustainable drop-in replacement for developers. Additionaly, once the zeal fades or the bubble pops, forced injection of AI into every site, app, and project will likely abate.
Where AI is Overkill
After searching “When was the US founded?”, an AI overview gives me three unique dates (1776, 1783, 1787) with descriptions of the relevent event for each and a Reddit link noting that the community generally agrees on 1776 as the ‘universally recognized’ year.
AI does not add value when retriving static information. “July 4, 1776”, the universally accepted date, is the only and correct answer. The rest is useless, although knowing that Reddit is the ultimate source of truth for is amusing.
Generative AI is fundamentally probablistic, if anything, it actually adds noise with queries like this. Tasks like document conversion, math, and retrieval gain nothing but noise.
Where AI is Friendly
A core part of development for many involves consultation of the code and process with an inanimate object, most commonly a rubber duck (example below). Talking through and explaining processes, bugs, concepts, and goals helps us understand them more deeply and this act often results in solutions and insights.
AI can be a great thought partner for this offering ideas or suggestions for things to check, areas to expand, or common standards for better code.
Note: AI does not, and should not, replace any rubber duck as the duck remains a better listener.
Where AI is Helpful
Dynamic Reference
AI truly shines as a dynamic reference and resource. When working on a project, especially if it’s a new method or content, AI can quickly generate ideas and a mental model to build on.
- For projects:
- How can I modularize this process?
- Any common naming or file standards?
- Can you give me a great example of this implemented? (Ask for link to existing, not generated)
- For libraries:
- Which are popular and industry standard?
- Are they actively maintained?
- Where can I find tutorials and examples?
- For analyses:
- How can I visualize this?
- Are there complementary approaches?
- Any assumptions or caveats for this?
Templates and Examples
AI also excels at example code, stubs, templates, and debugging. It’s risky to rely on AI to build a whole website, but it can accelerate the process when used to set up the framework and provide the base code for individual pages. This modular approach to utilizing AI quickly generates the necessary code base for a project, but avoids issues such as unexpected or unknown changes that can happen when it refactors an entire code base. It keeps the dev in the loop in terms of structure and overall project design but allows AI to handle much of the workload as well.
For short snippets, AI can also help since it is familiar with custom variable names, file structure, and setup. It acts as a quick reference when debugging or extending work. By only requesting short code at a time and integrating it manually, no unexpected refactors occur and does not drain tokens at an alarming rate.
Where AI is Risky
Over reliance on AI is exceptionally risky in several contexts. It can be wrong, offer responses meant to appease the user more than target appropriate processes, and fail to understand the big picture.
Many stories have emerged of vibe coded projects and businesses gone awry including non-protected databases, revealed keys, and unencrypted PII. When dealing with sensitive information, reliance on AI is utterly unwise. Security is tricky and ever changing. Without a proper understanding of the risks, AI may not protect your data, make recommendations that leave you open to attack, or bypass best-practices to just “make it work.”
Finally, if you have no role in the development and design of a code base then when things go wrong you have no power to fix or change things. “Built by AI” often infers it must be “maintained by AI.” Knowing the structure and patterns of your own work is essential to addressing issues and ensuring ongoing functionality.
Where AI is Heading
AI is a tool but a potentially dangerous one. Developers, at least presently, should absolutely use and leverage it but not depend on it entirely. Programmers will be still be valuable moving forward and the best way to ensure relevance in this new world is by understanding patterns, systems, best practices, and risks. AI is here to stay, but perhaps in a less invasive and ubiquitous way. Generative probabilistic responses are wasteful in many situations but offer quick and powerful aid in many others.
Don’t allow articificial intelligence to replace your own, only to supplement it.