Skills that keep engineers relevant in the AI era
- YT :: https://www.youtube.com/watch?v=siE6E0yVe7k
- Original title :: Don't Get Laid Off by AI (From a Lead Principal Engineer)
Generative AI is a fire-or-electricity-level shift, not just an incremental tool upgrade. If your job today is "receive context and a task from someone, then produce an output," you are already replaceable — AI does exactly that, thousands of times faster and without complaining about work-life balance. The video walks through how large language models actually work and then draws out three skills that let engineers benefit from AI instead of being displaced by it: acquiring context, staying versatile, and shifting away from pure craftsmanship toward business outcomes.
How large language models work
A large neural network is trained by reading effectively the entire internet, absorbing books, forum posts, code, and more into what the video calls "an empty brain." Once trained, it answers prompts that combine context and a desired task. Several limitations were then patched with bolt-on solutions:
- Stale knowledge (training cutoff) is fixed by giving the model real-time internet browsing.
- Weak math is fixed by giving it tools/plugins, like a calculator.
- Weak performance on specialized professions (teacher, lawyer, doctor, therapist) is fixed by fine-tuning the base model on domain-specific data, producing many specialized models.
- Limited context window (a prompt can only be a few thousand words) means it can't summarize a 1000-page book and has no memory between conversations. This is addressed with vector databases acting as long-term memory the model can query.
The key diagnostic: if your job matches this diagram — someone hands you context and a task, you consult external sources (Google, Stack Overflow, ChatGPT), you use tools (VS Code, Excel, Python) to produce an output — then your job is exactly what LLMs are built to do.
Whose jobs are safer
The real question isn't "which jobs are safe" but "what skills let you benefit from AI rather than be replaced by it."
Acquiring context
LLMs are good at solving well-specified problems but bad at acquiring context themselves. Given an exact file and function to optimize, AI performs well; given a vague brief like "our app feels slow," it fails, because someone first needs to profile the app, observe real usage, and figure out what "slow" even means before a solvable problem exists. People who can gather ambiguous context and turn it into a well-defined problem are far better positioned than people who wait for others to hand them fully-specified tasks and simply execute. This requires shifting from a task-driven mindset to a problem-driven, business-driven one. If your current role is to receive context and task from others, that's a warning sign to change now.
Build versatility
The more narrowly specialized your job title ("I'm a React developer," "I'm an Excel data analyst"), the more exposed you are. Fine-tuning a model to be excellent at one narrow task is comparatively easy and cheap — entire startups exist to do this and undercut human labor on cost. Building a model that's broadly excellent across many domains is much harder. Engineers who are versatile and can connect ideas across multiple domains — full-stack rather than locked into one language/framework/stack, willing to take on non-engineering tasks — are harder to replace. The goal is to be the person directing an army of AI models, not the person AI is slowly outcompeting at a single narrow skill.
Craftsmanship vs. business outcomes
The value placed on pure craftsmanship will keep declining, similar to the Industrial Revolution: hand-carved-chair carpenters lost out economically to entrepreneurs who built factories, even though factory output never matched handmade quality — because factories won on speed and cost at scale. AI will increasingly write scalable, maintainable, performant code faster than humans, even if it isn't the most elegant. Craftsmanship still matters, but engineers who define their pride and identity purely around it are at risk. The recommended shift is toward the entrepreneur archetype: understanding what makes a business succeed, taking on business accountability, and orienting toward business outcomes rather than technical purity alone.
Bottom line
To stay valuable as AI improves: get good at extracting and defining ambiguous problems (not just executing defined tasks), stay broad rather than narrowly specialized, and grow business/outcome awareness alongside technical craft.