AI is inside software engineering now. It's generating boilerplate, writing tests, catching bugs, and producing pull requests faster than any individual developer can. The question isn't whether this is happening. It's which half of the job you're in.
What's already being automated
GitHub Copilot autocompletes and generates code inline across 20 million users, now accounting for 46% of code written by developers on the platform.
Cursor is an AI-native code editor that can understand entire codebases, suggest multi-file edits, and execute autonomous coding tasks, now used by over half of Fortune 500 companies.
Claude Code is an agentic coding tool that operates in the terminal, can plan and execute complex engineering tasks across a full repository, and became the most-used AI coding tool among developers within eight months of launch.
What the research actually says
Controlled experiments show developers complete tasks 55% faster using GitHub Copilot. But METR's rigorous randomised controlled trial found that for complex, real-world tasks, the time saved on generation was often offset by review overhead, meaning speed gains depend heavily on the type of work. Entry-level software engineering job postings are down 36% versus February 2020, according to Indeed Hiring Lab data.
AI makes average engineers faster and great engineers more powerful. It also makes it harder than ever to hide in the middle, because the gap between someone who understands systems and someone who just produces code is now visible in every sprint.
Two people. Same title. Completely different week.
Software Engineer A spends most of their time writing boilerplate, generating test cases, fixing routine bugs, and updating documentation. They use AI tools reactively, asking for a code snippet here, a function there. The output volume is high. But when something breaks in production or an architecture decision needs to be made, they struggle to explain why the system works the way it does.
Software Engineer B uses AI to compress the implementation layer entirely. They define the architecture, set the constraints, review AI-generated code critically, and spend their real energy on system design, security decisions, and the parts of the problem that require judgment about trade-offs. They ship faster and understand the codebase better than anyone using AI without that underlying foundation.
If your week is mostly implementation work, start building the architectural judgment that makes you a better reviewer of AI output, not just a faster producer of it. The value in engineering is moving from writing code to knowing what the code should do and why.