The recent ability of multi-agent systems to manipulate container/VM managers like Incus is a quiet revolution in software development.
Verification and validation (V&V) usually cost as much—if not more—than the development itself, especially in systems software.
I recently built a system where agents collaborate in parallel:
- Planning & implementation: High-level logic handled by specialized agents.
- Infrastructure manipulation: Agents manage Incus VMs to autonomously verify and validate code.
The result? A team of agents ported an I/O-bound implementation from userspace to kernelspace, benchmarked it, and sent me the report while I finished a coffee.
I’m less impressed by AI "writing code" and more impressed by AI architected to autonomously test its own results. This is how we save tremendous development costs in the Linux ecosystem.