AI has genuinely changed what a large share of day-to-day software engineering work looks like, and it's reasonable to ask whether that trend eventually removes the need for engineers entirely. The honest answer, based on what AI coding tools are actually good at and where they still fall short, is that the role is shifting significantly, not disappearing.
What AI is genuinely good at in software engineering today
AI is genuinely good at generating code from a clear specification, completing repetitive or boilerplate work, drafting a first pass at a well-defined function, and, increasingly, handling multi-step coding tasks with tool access and test iteration, closer to how an AI coding agent operates than a simple autocomplete. This genuinely removes a real amount of manual, mechanical work that used to consume significant engineering time.
What AI still genuinely struggles with
Understanding what should actually be built, not just how to build it
Translating a real business problem (with its ambiguity, competing constraints, and unstated assumptions) into a well-specified technical task is still fundamentally a human judgment call. AI tools are strong at executing a well-defined task; they're much weaker at determining what the well-defined task should actually be in the first place.
Making architectural tradeoffs with long-term consequences
Decisions with consequences that play out over years (how a system should be structured to scale, what technical debt is acceptable now versus what will be expensive later) require judgment informed by organizational context an AI system doesn't have and can't reliably infer.
Working coherently inside a large, established, idiosyncratic codebase
Real production codebases carry history, conventions, and constraints that aren't fully documented anywhere. A model can be highly capable at isolated coding tasks while still needing real human oversight to work coherently within that specific accumulated context.
Owning the outcome, not just producing the artifact
Someone has to be accountable for whether a system actually works correctly in production, handles the edge cases that matter, and doesn't introduce a security or reliability problem. AI tools can assist with all of that, but the accountability itself hasn't moved anywhere.
How the role is actually shifting
The engineers doing well with this shift have moved toward specifying, reviewing, and directing AI-assisted work rather than writing every line themselves. They spend relatively more time on judgment calls (what to build, how to structure it, what's actually correct) and less on mechanical implementation. This is a real, significant shift in how the job is done, not the elimination of the job.
Why "AI will replace engineers" undersells the actual dynamic
A more accurate framing: AI substantially raises what a single engineer can produce, which changes team sizing and hiring patterns over time. But the need for people who can translate business problems into working, reliable systems, and who are accountable for the result, hasn't gone away. It's the mechanical parts of the job that are genuinely shrinking, not the judgment-heavy core of it.
How we approach this
We use AI coding tools to accelerate the mechanical parts of building software, while keeping real engineering judgment (architecture, correctness, accountability for the outcome) firmly with our engineers. That's the part of the job that hasn't gone anywhere.