What's Actually Changing Right Now
- AI-assisted coding is mainstream. Tools like GitHub Copilot, Cursor, and Claude Code are now a normal part of many engineers' workflow, drafting boilerplate, suggesting completions, and even writing whole functions from a description.
- The bottleneck is shifting from typing code to reviewing and directing it. More of an engineer's time goes into specifying what to build clearly, reviewing AI-generated output for correctness, and catching subtle bugs the AI introduced.
- New categories of engineering work are growing. Building the AI features themselves — RAG systems, agents, LLM integrations — is a fast-growing category of engineering work that didn't exist a few years ago at this scale.
What Isn't Changing (Yet)
- System design and architecture decisions still require human judgment about trade-offs, scalability, and long-term maintainability that current AI tools can't reliably make alone.
- Understanding the actual business problem — talking to users, figuring out what's actually worth building — remains a deeply human skill.
- Debugging genuinely novel, complex issues across a large, messy real-world system is still something engineers do better than AI tools, especially when the bug spans multiple services or has no clear reproduction steps.
- Accountability. Someone still has to be responsible for what ships to production — AI tools don't carry that accountability.
Which Engineers Are Most at Risk?
The engineers most exposed to disruption tend to be those whose main value was writing large volumes of routine, well-specified code — the kind of work AI tools have gotten genuinely good at. Engineers whose value comes from system design, judgment under ambiguity, cross-team communication, and understanding the "why" behind what's being built are, at least for now, much harder to substitute.
What Skills Matter More As a Result?
- Reviewing and directing AI output — knowing when generated code is subtly wrong, not just whether it runs
- System design — how pieces fit together, not just how to write one function
- Applied AI skills — understanding how to actually build with LLMs (RAG, agents, evaluation), since this is where new engineering demand is concentrated
- Communication — explaining technical trade-offs to non-technical stakeholders is not something AI tools do on your behalf in a room
- Judgment about when NOT to use AI — knowing which problems genuinely need a human-designed solution
The Honest Bottom Line
"Will AI replace software engineers" is, in most credible analysis, the wrong framing. A more accurate one: AI is replacing certain tasks within software engineering — largely the more repetitive, well-specified ones — while increasing demand for engineers who can design systems, direct AI tools effectively, and build the AI-powered products companies now want. The engineers who invest in those adjacent skills, rather than resisting the shift, tend to come out ahead.