The Future of Software Engineers
Several weeks ago I made a prediction regarding how AI will impact companies in the future. Here I have a prediction focused on which transformations AI will bring to software engineering as a profession.
It's a less serious prediction, but I want to write it down anyway because I like this "science fiction" scenario I made. I think it covers a 3-5 year horizon.
First of all, read this article: AI and the ironies of automation by Uwe Friedrichsen. I'm very aligned with his thoughts and my current prediction is a continuation of the ideas from the article.
Stage 1: AI Adoption is Skyrocketing
As of Q1 2026, we are at this stage.
The majority of big tech companies force the use of AI more and more. It leads to risks described in Uwe's article and risks described in my previous article. But right now things look bright, AI speeds up development unprecedentedly, AI-optimists are very excited.
The cost of a prototype has become super-low. Additionally, usage of LLMs requires leadership-like skills to get the most out of it (see Uwe's article). Considering this and the recent amount of layoffs, a plethora of startups are born: many (ex)managers and team leads can use LLMs to create pretty functional prototypes and gather teams around them due to their leadership skills.
Stage 2: Mittenspiel
Engineers in the big companies can be roughly split into two groups at this moment:
- AI-powered builders: They juggle multiple agent sessions across multiple git worktrees. They don't write code manually and their engineering skills are diminishing. But their delivery speed is very high, so stakeholders appreciate them greatly.
- Old-school engineers: They regularly write code by hand. If they use AI, they use it as a tool that empowers their skill of writing and designing code. They are usually unhappy with the LLM hype that is happening, but do not feel safe to explicitly say it to the management. To save their jobs they can mimic AI-powered builders. Regular practice of writing code by hand helps them to use LLMs effectively enough to save their jobs.
In the most extreme cases, the second group can be severely punished for not using AI enough.
Existing big software product contains architectural decisions, patterns and practices, huge array of docs and historical records of decisions and discussions. Having autocomplete properties, LLMs always try to follow them. This postpones the moment where damage from excessive LLM usage is non-negligible and LLMs cannot properly deliver features anymore.
Vibe-coded startups have no such inertia and will reach their limit much faster. In such startups AI-powered builders will reach the point of despair sooner and they will need the help of "AI fixers" earlier (see Uwe's article for definition).
Stage 3: A New Purpose
Old-school engineers preserved their skills and are currently the best AI-fixers on the market. Naturally, startups that are looking for AI-fixers will become very attractive for them, considering that demand from big companies is not yet fully emerged. Startups' leaders will recognize that old-school engineers must be in the team in order to have a great product.
So, old-school engineers will leave big companies and join startups, where they can take a more meaningful role and where their skills and personality are more appreciated. Considering the layoffs, no engineer will perceive working in big company as a safe bet anymore.
Stage 4: The Paradox
Big companies reached the point where they need AI-fixers, but they lost many of them and are not attractive to them. So they will pay huge money to motivate more AI-fixers to join them (or to train their AI-builders to restore required skills). But considering the size and complexity of big products and the amount of LLM-generated code, reverting the damage from excessive LLM usage will be very time-consuming.
At the same time, many AI-fixers already helped startups to build balanced and sane approach to AI usage. They already fixed major issues and made a synergistic relationship with AI-builders. Failure of big companies expands opportunities for startups and makes them more attractive for investors.
At this moment:
- Those who bet everything on AI will suffer from AI the most.
- Those who lost faith into promised AI miracles early enough will benefit from AI the most.
Conclusion
If you want to make God laugh, tell him your plans.
Nevertheless, I believe that keeping ability to write code by hand is a good bet for any software engineer in the scope of the next 3-5 years.