How Will Software Engineers Lose Their Jobs Within the Next 5 Years?

Approximately three years ago, I was chatting with a dear friend on LinkedIn and predicted that in around 5-10 years, we will have advanced systems that will generate code automatically and there will be only IT managers that will only say what to do in human language.

In this article, I will try to explain the current situation and offer insights on how to ride the AI wave effectively.

The Situation Is Critical if You Never Heard About CoPilot, ChatGPT, and Other Popular Tools.

A long time ago, with new intelligent systems like IntelliSense, coders, who started to use new technologies and didn’t know all available methods of the frameworks, have become the butt of jokes among seniors.

They were complaining about how a developer doesn't know all the methods that technology has.

Today, it is quite challenging to write any code without autocomplete systems, especially on the whiteboard during interviews, because we all get used to it.

To my big surprise, many software engineers, especially those who became grandmasters in their spheres, are not interested in new technologies at all, especially, since they believe that AI technologies don't intersect or interact with coders.

Experts are people who are not interested in knowing new things because if they do that, they will not be experts; paradoxical, isn't it? It was true when the world saw the first good results in image recognition, google translate, data analytics, etc.

But, there were no good results in language generation. Now, things have changed a lot with GPT-based technologies, because basically, coders are writers.

CoPilot and ChatGPT Are Your Pair Programmer?

Today, not so many coders are using these tools in daily work, some of them prefer old-style coding styles, and some of them can not use them because of their company’s privacy policies.

The idea is simple; in most daily work, programmers have routine repetitive activities such as transferring one data to another data, writings tests, and other low-brain costs repetitive work.

Github CoPilot was created to help programmers concentrate on more important tasks, generating code for routine tasks.

CoPilot is GPT-like Large Language Model (Generative pre-trained transformer). The basic idea is that it tries to predict the next word in a sequence of words many times as needed until it understands that it is enough. How do you use that?

The coder should buy a subscription, install the CoPilot extension for IDE, and write something in human manners inside IDE, and CoPilot will generate the answer to it (Img 1).

You can write the name of the function, specify input data, write some comments on what the function is doing, and CoPilot will generate the body. Something very similar I did when I was working for a big company.

Senior developers write a project template, the architecture, specify the name of functions, describe what should be done in comments, and also, add some input and output variables. Junior developers write some code inside functions, and middle developers help them to verify everything.

We have a whole big department of programmers that were working like a big organism, creating RPA bots in a Ford conveyor manner.

The first time I tried GitHub CoPilot, I felt the same feelings. Yes, it generates the code and the quality was higher than our junior developers were writing that time, but it still has the same problems that I had working like that. You need to read another's code and understand it.

Understanding is important because you need to fix this code by yourself or rewrite it by creating new prompts for CoPilot. Through iteration, you will make this code work fine, and a big advantage of communicating with AI is the speed.

You don’t need to wait a lot, while junior developers will think about how to do that; also, AI knows a lot of coding languages, so you have a Swiss knife in your pocket.

Finally, you get a good assistant for your work that speed up your coding session but not a fully independent AI coder.

ChatGPT is a GPT-like LLM that also can be integrated into your IDE using OpenAI API. While CoPilot is designed to generate code, ChatGPT is more general and has the extra ability to answer any questions, which also includes the coding sphere, which will be interesting to create new technologies…

What’s Next?

The main problem is we still need to interact with ChatGPT via prompts, read many codes, and spend some time fixing bugs, etc. Autocomplete is outdated, modern GPTs still have problems, What do we have now?

AI-Agents, Auto-GPT

AI-agent(s) is an agent, that replaces you in this chain of events. Instead of you, they interact with ChatGPT and other GPT-based technologies via APIs. This technology tries to fix the main problem that we already have described. The current modern technology, for now, is Auto-GPT (Img 2).

Auto-GPT is an experimental open-source application showcasing the capabilities of the GPT-4 language model. This program, driven by GPT-4, chains together LLM "thoughts" to autonomously achieve whatever goal you set.

The basic idea is an action chain that has a “Observer, Decide, Act” loop.

You just need to set up the goal and wait some time. It will automatically interact with GPT via prompts and evaluate the goal achievement. The technology is not optimized for now, and we can get endless loops where agents will do prompts so many times, but the goal will never be reached.

Also, each epoch will cost some money because you need to pay for each prompt via API, but still, technology is the start of the sunset of a classic programmer's career…

LangChain and Transformers-Agents

LangChain - trend framework technology that came after AI agents. Based on LLM, LangChain contains many LLM chains inside and becomes a solid, completed product. LangChain supports AI agents, different APIs, custom functions, etc.

Transformers-Agents - a product that extends AI-agents and LangChain further more. The basic idea is to choose what AI product to use automatically because we have many of them, and the problem is which is better for a specific task.

Using this technology, we can not only generate code but also create designs or generate voice, etc in different spheres.

Hugging Face is a company (Img 3), that stores different neural networks, various AI products, datasets, etc. all in one place, and Transformers-Agents can intellectually choose what AI products to use, depending on your input goal.

Final Thoughts…

We must progress and stay updated with the evolving trends in the software engineering domain. My prediction is that finally, in the near future, traditional programming roles will diminish, leading us to pivot to roles like IT management.

Coders will lose their jobs and start to be a bridge between cutting-edge AI code-generating products and businesses that commission this software. We are living in an interesting world.

Thank you for your attention!