Open source devtools debate gets sharper and messier
The 'Devtools must be open source' thread generated real friction. The original premise is that devtools should be open by default so users can examine and modify behavior. The pushback is equally pointed: most users, even expert programmers, never actually do this, and the argument breaks down further when you factor in AI agents customizing tools on your behalf. Who is responsible when an agent modifies a tool and something goes wrong?
The Pandoc 20th anniversary thread ran alongside this and offered a quiet counterpoint. Pandoc is beloved, has been maintained for two decades, and represents what durable open source tooling actually looks like. The comments were genuinely warm in a way that is rare for HN.
The pattern connecting them: the open source devtools argument is being complicated by the agent layer. When a tool is configured by an AI agent rather than a human, the classic 'freedom to inspect and modify' argument shifts. The liability and trust questions get murkier, not clearer.
So what?
If you are building devtools, the agent customization question is not hypothetical anymore. You need a clear answer to 'what happens when an agent misconfigures this and causes damage.' Open source does not automatically solve that, and closed source does not automatically cause it. Design your tool's interface and permissions with agents in mind, not just humans.