Look back at previous technology advances to make sense of this AI thing
How will AI change software development?
There are a lot of hype everywhere. It’s hard to judge how AI will change education, legal and health-care. Just like you, I would like to have better clarity on what’s real vs what’s just speculations.
So, I immersed myself into these AI tools for a few months and build a few apps to get a better feel on how it would affect software engineers.
Here are my responses to some FAQs. Hope you find them useful.
If AI is so good, should anyone still learn computer science or pursue software engineering as a profession?
AI tools are very useful accelerators.
It’s good at performing well-defined tasks quickly: Crafting UIs, managing data, examining reams of logs to find errors and suggest solutions to them.
An experienced engineer can now direct AI to finish these type of work in 1/4 of the time it would have taken them to do them the old fashion way (pre-2025.)
Today’s coding/debugging work will become easier . That part is very clear.
The non-coding parts of creating a product (e.g. figuring out what to build, arriving at consensus, validating the user experience in real life, looking at the usage information to refine the product, managing/marketing the project.) will benefit from AI, but less dramatically.
Software powers the world and will continue to do so. The need for good software creators will not go away because of AI.
Having said that, the bar is changed. All good software engineers will be well-versed in using AI tools, just like traditional tools such as debuggers.
In short, if you enjoy creating software, there is no reason to avoid pursuing it as a profession. It will continue to sustain a rewarding career. AI will not change that.
How can I adapt to this wave of changes in the profession?
If your organization has started rolling out these new tools (like Copilot, Cursor, Claude Code, Gemini Code Assist), immerse in what’s available to you. Those who can use these AI tools effectively will be ahead of the pack. That’s also very clear. It’s nice to be able to ride a bike and a motorcycle.
If your organization is still on the fence, start a side project to pick up experience on these tools. It will pay off in your current role or future roles. Hosting web apps is practically free nowadays. Getting a mobile development account from Apple or Google in order to publish an app is not very expensive. It’s a high-leverage investment of time.
But I don’t know where to start?! Here is the interesting twist. Just pose your question to Gemini / ChatGPT with some description of what you aim to do. They do a decent job on listing out the steps involved. All you need to do is ask to get started. If you prefer a human, email me.
What do I need to learn this stuff efficiently? Books? Powerful laptops? Nvidia graphics cards? AI subscription like Gemini Pro / ChatGPT / Claude / Cursor / Copilot?
In my humble opinion, from a return-on-investment viewpoint:
Cursor $20/month is on top of my list. Free for students.
I prefer Gemini Pro over ChatGPT Plus and Claude Pro. I have much data on Gmail / Drive already. This way, the data loss risk is lower because the attack surface is smaller. Gemini Pro and ChatGPT Plus is free for US/Canadian students as well. I pay for Gemini Pro version because I find myself being slowed down by the free version’s limits, plus I get more Google storage for other things.
Powerful laptops or Graphics Card? It’s really optional. Yes, you can learn a bit more about local hosting of LLMs. When I need big powerful machines, I would just run the code on cloud providers (e.g. ML training jobs). Way cheaper than using personal hardware if you only use it occasionally.
A good middle-of-the-road solution is an Apple Silicon Mac with 16Gb+ RAM. It can run local LLMs without breaking the bank. It’s cheaper than buying Windows PC with video cards with 16Gb+ VRAM.
Books? The field is moving too fast. I find the answers from chatbot good enough.
How is AI similar to and different from previous advances?
After discounting for recency bias, here are a few highlights:
Like previous advances like the Internet, AI doesn’t remove the need for strong foundational software engineering mindset: Clear logical thinking, ultra-patience, methodical, comfortable with details, willingness to abandon the obsolete when there is something better.
Unlike previous advances, AI is very uncanny. We still don’t fully understand how it comes up with a specific response. There are simply too many layers of processing for any human to be able to step through it and explain everything.
Unlike previous advances, AI gives us different outputs even if the same input is fed into the same software. It’s confusing for logical thinkers. New techniques are needed to account for this.
What are my next steps?
Work on my critical thinking skills: Re-read literature on this topic.
Keep writing: It helps me understand things better. System 2 thinking.
Explore how to use AI to better manage my investments.
That’s all for now. Until next time…