Accountability for what ships
90Someone has to be answerable when production breaks, a security hole leaks data, or a system makes a harmful decision — a model cannot hold that responsibility.
Software engineering is one of the first professions on this site to feel large-language-model automation directly, because its raw material — text that follows strict rules — is exactly what these models were built to produce. Tools that write, complete and review code are already part of most professional engineers' daily workflow, not a future scenario.
That does not make the job obsolete; it makes the valuable part of the job narrower and more concentrated. The sections below separate what is already being automated from what has, so far, proven much harder to hand to a model.
High
Share of the work a machine could do
A large share of the code an engineer types in a given week — boilerplate, first-draft tests, routine translations — is already comparably fast for a language model to produce. What remains hard to automate is deciding what should be built, catching the ways a plausible-looking change can be subtly wrong, and being accountable when it fails in production.
Scored from the tasks, not the job title. Lower is safer.
Jobs AI cannot take →What machines cannot take
Translating ambiguous needs into a spec
83Turning a vague business request into a precise, buildable plan still requires asking the right questions of people who often do not know what they want yet.
Whole-system architecture
77Deciding how dozens of services, teams and data stores should fit together, and living with that decision for years, is a different skill from writing any one function.
Debugging failures with no precedent
74Novel production incidents — the ones with no matching forum post or training example — still need a human forming and testing hypotheses under pressure.
Cross-team negotiation
80Deciding whose team absorbs the cost of a shortcut, or which of two conflicting deadlines slips, is a political and technical judgment call at once.
What they already take
Boilerplate and glue code
82Wiring together well-understood patterns — CRUD endpoints, form validation, standard configuration — is now often faster to generate and check than to type by hand.
First-draft unit tests
70Assistants can read a function and produce a plausible starting set of test cases quickly, though someone still has to judge whether they test the right things.
Routine code review comments
60Style, naming and obvious-bug feedback on a pull request is increasingly left to automated linters and AI reviewers, freeing human reviewers for design-level comments.
Porting code between languages
65Translating a function or a small module from one language or framework to another is a pattern-matching task language models are already fairly reliable at.
How the work is changing
From typing code to reviewing it
A growing share of an engineer's day is spent reading, testing and correcting code a model proposed rather than composing every line from a blank file.
Smaller teams, wider scope
Teams that once needed several engineers to cover a product area increasingly expect one or two, leaning on AI tools to cover the resulting gap in raw output.
Precise specification becomes a core skill
Writing an unambiguous description of what a system should do — once left implicit in a senior engineer's head — is now a skill worth teaching directly.
Verification becomes the bottleneck
Because generated code can look correct while being subtly wrong, testing discipline, code review and monitoring are gaining status relative to raw code production.
New jobs branching off
AI / ML engineer
Builds and fine-tunes the models themselves rather than the applications around them; overlaps heavily with traditional software engineering but leans more on statistics.
Prompt / context engineer
Designs the retrieval, tooling and prompt structures that make an LLM-based feature reliable enough to ship, a discipline that barely existed before the early 2020s.
AI red-team / safety engineer
Tests models and AI-written code for security holes, bias and failure modes before they reach production, an emerging specialisation inside larger security teams.
Site reliability engineer (SRE)
Already existed, but is gaining relative importance as the accountable human for systems that increasingly write, test and even deploy themselves.
Outlook
Software engineering will very likely keep paying well and stay in high demand through the next decade, but the shape of an entry-level job is already changing: fewer roles will exist purely to produce routine code, and more will expect a junior engineer to review, test and take responsibility for AI-generated work almost immediately.
The advantage will increasingly sit with engineers who can frame problems clearly, reason about systems as a whole, and catch what a fluent-sounding but wrong suggestion gets wrong — the same judgment that mattered before large language models, now applied faster and at greater scale.
Keep exploring
More in Engineering & Technology
🤖
AI Researcher
Designs and tests the algorithms behind machine intelligence, in a field now racing to automate a growing share of its own research process.
AI-resistant 50 🛰️Aerospace Engineer
Designs, analyzes and certifies the aircraft, rockets and spacecraft that leave the ground, working to safety margins that leave no room for guessing.
AI-resistant 74 🔌Semiconductor Engineer
Designs and fabricates the transistors inside every computer, phone and weapon, using machines precise enough that only a few factories on Earth can run them.
AI-resistant 60