Earlier quoted context omitted.
The claims being made here are fallacious in three extremely basic ways. First, "I regularly have 10+ instances of my harness running or idling on my laptop" implicitly asserts that you need that many instances running at once. You don't. You can have a single instance running that handles your sessions in parallel. Second, "I sometimes run my harness on machines that are low-power, low-RAM single-board computers." p…
This is getting silly, I can’t tell whether you are kidding, never launched a Node-based app, or never checked htop. > that can run their tool calls in a separate process/host That indeed seems like a simple and elegant setup. I can’t recall a single instance of using a harness when I didn’t want the tool calls to be local, though. Why would I be SSHing into a SBC just to run the agent back on my laptop? > You can ha…
You're very much operating based on emotion rather than reason. That's fine in general, but it's not OK for HN, which is not a place to blast out your emotions and try to assert that your opinion is factually true by getting maximally angry about it.
> That indeed seems like a simple and elegant setup. I can’t recall a single instance of using a harness when I didn’t want the tool calls to be local, though. Why would I be SSHing into a SBC just to run the agent back on my laptop?
You don't understand how harnesses work. Let me explain: in a harness, the tools that the harness provides are the only way for the agent to interact with its environment, and as such, if the harness is run on one host, and the tools on another, the agent's perspective is that it is running on the second host. This is effectively true modulo some small set of scenarios that are mostly irrelevant in the case of "I'm SSHing into an SBC" (e.g. the harness logic and LLM API calls would both be happening on the SBC, which you definitely don't want when you're resource-constrained).
> Sounds even more elegant! Presumably, they communicate via HTTP?
No. Why would you think that? Again, I don't think you understand how harnesses work.
> I don’t need to, but I can. That’s the entire point, in case you’re still missing it.
That's a useless point, if so, akin to saying "I want to have Vim running at address 0x00000001beef0000 in physical memory" - you're conflating use-case (needing a large number of agents running in parallel so you can work on things in parallel) with implementation detail (having a large number of separate processes running on your computer).
> How beautifully ironic.
I'm meeting every single one of your technical arguments and factual inaccuracies (e.g. your misunderstanding of the word "contrarian"), and you're continuing to get very angry at Javascript. I don't think there's any irony here.