Can AI Take a Software Engineer's Job?
I had an interesting exchange with a coding agent this evening.
Are coding agents useful? Absolutely. But can it take a software engineer’s job? You tell me.
Me: In .gitlab-ci.yml, can we set the workflow name only if the PLAYBOOK variable is set?
AI:
| |
The
workflow.nameonly takes effect when matched by a rule, so on push pipelines (where$PLAYBOOKis empty) it falls through towhen: alwayswhich uses the default pipeline name. When$PLAYBOOKis set, it shows e.g. “shape playbook”.
Me: You just made that up.
AI: You’re right — GitLab CI doesn’t support conditional workflow.name like that. The name field is always evaluated regardless of rules.
There’s no way to conditionally set
workflow.namein GitLab CI. It will always show the interpolated value. When$PLAYBOOKis empty, it’ll just show “Playbook “.Not much we can do here without GitLab adding support for it. Leave it as-is?
Me: What if we set a variable conditionally? And then use that variable as the name here…
AI: Good idea — use workflow.rules to set a variable based on the condition:
| |
When
$PLAYBOOKis set, the pipeline is named “Playbook shape” / “Playbook update”. Otherwise it falls through to the default.
The AI confidently presented a made-up behaviour, admitted it when called out, then declared the problem unsolvable, until I suggested the approach that actually worked. It wrote the implementation, but the solution was mine.
AI coding assistants are powerful tools. But “tool” is the key word here.
comments powered by Disqus