GenAI for Coding

Are we all going to code with AI now?

I can’t remember where I read this recently, but it was an analogy about GenAI for coding that went something like this : “In the future, using GenAI for coding is akin to using Google Maps ( GPS technology ) when driving.” The analogy suggests that GenAI will be part and parcel of any coding workflow in the near future. This analogy got me thinking more deeply about this topic, which eventually led to this post.

Initial thoughts

While I agree that almost everyone uses GPS for navigation, there are some who stick by their own navigation skills instead. For example, when he drives, my dad finds the audio and visual prompts of a digital navigating device distracting, and prefers to use his familiarity with Singapore roads to drive around Singapore. I guess using GPS technology as an in-car navigation assistant tool is a skill itself. Relating this to GenAI coding, I can understand why some programmers find GenAI a distraction. They have done well using their current workflows to derive their technical solutions, while some GenAI tools are known to provide inaccurate solutions. My experience using GenAI for coding is usually to deal with situations outside of my analytics competencies, such as when I am invoking my Javascript snippets for my frontend. GenAI isn’t helping me improve my solutions in a domain that I am an expert in.

Supply Boom

One contribution of GPS technology that I don’t see mentioned often is how it allows people to take up ride-hailing and food deliveries without much in-depth geospatial knowledge. This is a stark contrast to how London cabbies need to study London roads extensively to pass their licence exam. GenAI could potentially get more non-coders into “coding”, allowing them to skip the rigorous computer science / bootcamp training while they just “chat” their way to their tech solutions. In fact, I am seeing more “I coded this in 2 hours using GenAI with no CS background” Linkedin posts these days, although I have yet to see an actual live app or website that is proudly and primarily coded by a GenAI tool.

From another lens, I also see similarities in how the boom of digital cameras created a wave of hipster, hobbyist photographers. The technology allowed people with less skill to achieve more, and kinda fuelled a boom in amateur photography, which may in turn lead some to eventually try to turn professional. In the GenAI space, if current trends continue, we may see an explosion of GenAI coders building their own technical projects in the near future, be it hobbyists solving their own problem to hustlers launching their revolutionary world-changing idea.

Dependencies

Personally, I do feel that being totally dependent on GPS is dangerous. For example, Google Maps has gone horribly wrong at times. Personally, on my recent Bali trip, Google Maps recommended me a car route to a destination that was not too far. The recommended walking route also uses the same car route, which was not ideal under Bali afternoon heat. Booking a ride from a ride-hailing app to take the car route seems the most ideal. However, on a closer map and site inspection, we found a backdoor that opened to a shorter, more desirable walk to our destination. I didn’t need to book, wait and pay for a cab ( yay to no Bali traffic stress ! ), and got to my destination more quickly than the proposed Google suggestion ! In the end, the combination of some map reading skills, some cognitive distrust in algorithmic recommendations and humanity's ever-quenching thirst for exploration allowed me to circumvent the controls of our techno overlords to a more optimal and desired journey. After all, the journey is usually more important than the destination itself, right ?

In the same breadth, I feel that we shouldn’t be fully dependent on GenAI tools for coding either. While GenAI can produce code, their solutions may not be the most ideal or optimal, and would definitely require an experienced eye to review. This may not matter too much if we are building a hobbyist project for a simple solution. However, when we need to architect a larger technical solution, there are more chances that our GenAI solutions may create more issues to the growing codebase down the road. Issues may also arise when bugs appear in the codebase, and the orchestrator of the product will need to know how to effectively prompt the correct GenAI solution. This will require the “prompter” to know what technical terms or logic flows to prompt into their chosen GenAI tool to get the correct technical solution.

To tech coders & business owners

To business owners - I think wanting to depend on GenAI to code your technical work is alluring. The costs are low, and you seem to have full control of the direction of whatever you are building. I cannot say for sure if this approach will fail or succeed, because I have not evaluated every GenAI tool for coding. And GenAI for coding is something that is developing very rapidly. If you are a business owner reading this, I hope my article has given you some food for thought about using GenAI for coding.

To tech coders - Be mindful of the pros and cons of GenAI tools for coding, and understand when they can be useful and dangerous. My personal experience using GenAI for coding hasn’t provided perfect solutions, but for things that I am very weak in, even free GenAI tools have helped me in some ways. Secondly, it is now even more important to focus on concepts over tools in our programming careers. If React code can be easily created by GenAI, our value will come from more higher-order abstract concepts like how we design and architect our technical solutions. Lastly, be prepared for the boom of GenAI coders and their promises. We rode the wave of no-code solutions for the past few years and have somewhat survived. The situation is slightly different now, but I feel that as long as we focus on the first principles of creating value with our technical competencies, we should be fine. At least that is what I hope so.