Is vibe coding used anywhere in spacecraft software development, or is it effectively banned due to the criticality of the domain?
I'm curious how the space software industry (NASA, ESA, commercial companies, etc.) actually treats AI coding assistants (Copilot, Claude, ChatGPT, etc.) in practice. Specifically:
- Is there a formal or informal ban on using LLM-generated code in flight/mission-critical software?
- If AI tools are allowed, where exactly are they used — requirements analysis, test generation, boilerplate code, documentation — versus where they're strictly forbidden (flight control logic, safety-critical modules)?
- The main issue I keep hearing about is explainability: AI output can't be traced back to a clear rationale the way a human engineer's design decisions can, which seems fundamentally at odds with standards like DO-178C that require full traceability from requirement to code to test. Is this the actual blocker in practice, or are there ways around it (e.g. formal verification on top of AI-generated code)?
Would love to hear from anyone with direct experience in flight software, avionics, or satellite software development.