Software Engineers Survival Guide
Disclaimer: It’s been a while since I last updated this blog, and the reason is simple – I am Ukrainian, and the war is still ongoing. Living in Kyiv, I hear attacks on us almost every other day. But there’s another part of the story – my husband has joined the Army, and I am now managing the family business he ran for the past 13 years.
It’s a technical recruiting agency founded by former engineers. Over the past year, I’ve been conducting initial technical screenings for backend, frontend, and low-level developers, having numerous conversations with engineering managers, and learning a great deal about the hiring process for exceptional startups.
The insights in this piece are backed not only by my experience as an MLE and my daily use of Cursor but also by extensive hands-on recruiting experience.
We can see a whole spectrum of opinions: from those claiming we no longer need even mid-level engineers to AI-based agents will only create more bugs (and we're all going to hell but there will be old-school true engineers who will save us).
As it is often true - both sides have some valid points. Let's have the basics covered:
- SWE is a verifiable domain with a relatively quick feedback loop (the code is either working or not). We are not touching here if the code is optimal, doing what it was initially intended to do, or even if it is safe - these aspects are way harder to verify and require human oversight.
This feedback loop and the code generation itself is a nutritious environment for LLMs specifically. It means that AI agents/coding assistants/AI-coders will become better and better. Check out the progress made on SWE-bench – from 12.5% percentage of instances solved to ~65% in ~1 year. Yes, it is a just benchmark, real-world tasks are more sophisticated (irony sign here), and getting to 100% is actually not feasible.
- Some of us witnessed IDEs development, from clean text to highlighting syntax errors, filling names... Now they can index the whole repository, generate a bunch of code, fix itself, add docs, and... easily expose your creds to the world. But the main point still is – it will get better and penetrate in more domains. And yes, we will face crazy stupid bugs due to vibe coding and non-engineers using these IDEs. The world is complex.
As a software engineer, or a manager, or a fellow wanderer you might be a bit lost what to do and how to excel in these madness.
1) You have to become a waaay better engineer
- Improve your skills in system design - even with crazy long chains and context window systems design and fixing is still farther away
- Work on developing system and critical thinking. These two are different types
- Teach yourself to be able to navigate and understand any code quickly, develop and eye for the loopholes, system-level problems, inconsistencies
- Learn to effectively pair code with AI assistant
For instance below is a few categories which will definitely require more human-in-the loop even if AGI will be here tomorrow:
Complex system architecture and design patterns
Maintaining long-term code quality, structure, and scalability
Managing evolving tech debt and dependencies
Performance-critical applications and optimization
Real-time systems and low-latency code
Edge-case handling, exception management, and fault tolerance
Security-sensitive code (e.g., authentication, encryption, data protection)
Navigating compliance, regulatory, and legal constraints
Legacy system integration and modernization
Handling ambiguous or underspecified requirements
Deep domain-specific logic and complex business rules
2) You have to specialise
I'm not talking about backend, frontend, I mean some real-world domains – manufacturing, cybersecurity, defence tech, retail, health care, etc. The narrower the domain is, the better. Have you seen the software in some industries? It is pre-SAP, and these industries will continue living, but I would not say the same for endless flow of things trying to disrupt something by creating cutting-edge BS
3) You have to extensively work on your soft skills
Are you able to plan and set priorities for yourself? To pair code? To pair code with an AI assistant? To collaborate with your teammates? To resolve conflicts? To explain anything to non-tech people? To lead discussions? To give and receive feedback? To act on the feedback?
Sit down today for at least an hour and make an initial self-assessment/brain dump:
- What technical skills are you lacking? Both in depth and in breadth. For instance, do you understand how databases work under the hood? Or how ML is trained?
- In what domains do you have the most experience, and in which ones would you love to gain some? Why?
- What soft skills are you lacking? List a few use cases where you acted on negative feedback, or when you went the extra mile to get the results needed. This will also serve as prep for behavioural interviews.
Be honest with yourself.