Hacker News Reader

What kind of features are folks implementing that require a small “team” of parallel agents? I get the feeling a lot of this use case is for start-ups and/or folks trying to self-start a SaaS or something. Personally, there’s nothing at my job that I find needs more than maybe a prompt or short back and forth with a single LLM to accomplish, but then again the project has been established for about a year or so now.

View on HN

For backend stuff it's integrations. Say you have a solid core and you want to integrate with 3-5 3rd party tools. You can write one, and then do something like based on the example at @file and docs @docs, integrate provider $n.

For frontend stuff my colleagues experimented with "rolling the dice" multiple times for stuff that the models usually, but not always get right. Say a component with defined styles or conventions. You can see at a glance if it's right, but if it's not you have to go in, change the prompts, etc. Having the model roll a few choices and you just pick and choose is faster. Especially combined with fast and cheap models like fast-code-1 or gpt5-mini.

Reply on news.ycombinator.com