FAQs on AI replacing human software engineers
Yes. Some of them, especially those:
- Doing tedious and repetitive work.
- Executing well-defined tasks given by others.
- Building simple CRUD apps and websites.
Essentially if most of you daily work can be automated by hiring Devin ($500/month) at a cheaper cost, you are at risk.
No. Here are some cases where software engineers won't be replaced in the near future:
Specialized software
- People working on specialized software that are out of distribution (OOD) or don't conform to typical software patterns (OS kernels, DBMS, aircraft control, oil refinery management) are safe from AI because AI don't have enough knowledge on how to work on them.
Mission-critial software
- People working on mission-critical software like aircraft control or medical devices will probably be safe as there are strict regulations and less room for error.
Accountability & supervisory role
- Business still need some people to supervise the AI, review AI-generated code, and be accountable for the code. The ideal person to fit into this role would still be a software engineer.
I think the first wave of software engineers getting replaced or made redundant directly due to AI will happen in 2025.
In 3 years "software engineers" will become a niche profession that only exists in specialized industries, like how Perl or COBOL today.
I think of replacement in terms of number of human employees vs AI employees.
Hypothetically, a company currently has 5 human engineers. With an addition of an AI engineer, the company now only need 4 human engineers + 1 AI engineer to achieve the same output. If the company do not require more engineering bandwidth, then the company can decide to layoff one human engineer.
That human engineer is "replaced".
The AI engineer might not be able to do 100% of the things the human engineer can do, like updating JIRA ticket status and attending meetings. But we still have other 4 human engineers that can help with those tasks.
The AI engineer can just focus on things that it can do, like analysing the requirements, reading the existing codebase to understand its structure, writing the technical plan for the new feature, asking for clarifications, writing the code, writing tests, running the tests locally, and finally submitting a PR for review. Devin is one example of an AI engineer that can perform these tasks.
Check out my first impressions on Devin here.
The technology in the AI coding industry is advancing very fast, with multiple companies moving towards building L4-level AI software engineer.
I recently got access to Devin in January 2025, and I am very impressed with its current product offerings and capabilities. There are some constraints and silly issues but I think those will get sorted in no time given how fast things move.
Check out my first impressions on Devin here.
From the business perspective, there are major advantages of hiring AI instead of human software engineers:
- Don't need to deal with hiring at all. Instant hire, instant onbaording.
- Don't need to deal with ramping up the team size, hire as many as you want. Infinite scalability.
- Don't need to worry about productivity. AI works 24/7 and don't slack off (yet).
- Don't need to worry about retention. AI will not submit resignation letters, they will work for you until the end of time as long as you pay for them.
If you are still using GitHub Copilot and ChatGPT as an example of how AI can't replace software engineers, you need to keep up with new development in the past 6 months:
If you are using the raw model, use Claude 3.5 Sonnet. It crushes anything OpenAI released to public in the last few months. The only close contender is o1, but in my experience Claude 3.5 Sonnet is still the best for coding.
Use a L2 tool like cursor, aider, Cline or 16x Prompt to integrate Claude with your codebase context and only pass relevant context to the LLM, this dramatically improves the output quality and reliability.
Experiment with a L4 tool like Devin, Marblism or bolt.new to see how the trend is evolving in full stack end-to-end development. They are stil in the early stage, but might become better rapidly with new frontier models like o3.
No. There will be other professions born out of the AI adoption.
For example, software promoters who are technical people and specialize in improving effectiveness and efficiency of building software using AI tools. Like how sales engineers or solutions engineer today don't directly code the product, but help clients get the most out of the product.
Software engineers can prepare for the future by learning to use AI tools to be more productive. In this way, they can produce more value when paired with AI tools, compared to just AI tools like Devin alone.
It is also good to move up the value chain into product management or team lead (manager track).
As software engineering and code become more commoditized, the value creation will shift up the value chain and product management will become more important.
Team leads are still needed, regardless of it's human engineers or AI engineers that they manage. Someone needs to supervise them and review PRs. And someone needs to communicate with the business and product team to discuss priorities and schedules.
A common question is: "If we replace junior software engineers, where do we get the senior software engineers?"
The answer: The same way we have always been, through companies hiring them, and self-learning.
Large companies may continue to hire and train junior software engineers (in reduced numbers) to:
- make sure they have a pipeline for senior software engineers in case things go wrong
- fufill the social responsibilities of ensuring there are humans to pass down the knowledge (Captain B. McCrea in WALL-E)
Plenty of people who are passionate about coding will continue to learn programming on their own by reading books, following tutorials and building, with the help of AI.
I have worked at tech companies like Grab and Ant Group (Alipay) for 5 years (2017-2022). My last title was Senior Frontend Engineer at Ant Group.
I built AI Simulator mobile games in 2022 with PPO model training and inference running on device (and fixed some bugs in TensorFlow.js along the way).
I built 16x Prompt to streamline AI coding workflows in September 2023, been using AI to code almost daily for a year, acquired 300+ paying customers.
I replicated GPT-2 results myself on Azure GPU instance in December 2023 with nanoGPT.
I made over 3000 contributions to 40 repositories in 2024, 94% of them are commits.
I have consulted with several startups as fractional CTO and software consultant.
Here is a post where I discussed why I am qualified to make such predictions.
Here is my LinkedIn profile if you'd like to find out more about my experience.