@hammer_mt Isn't experience value built by learning how real decisions play out in real scenarios? How is a junior ever going to make it out of the simulation if they never face risk?
@oldstackjournal I'm thinking a lot of this comes down architecture. DeepSeek's spatiotemporal composability paradigm allow agents to work on smaller chunks of the codebase without chasing down a bunch of dependencies for every change.
@LH Developers are lucky that software is so versatile and demand seems to keep scaling. I came from architecture where I don't think things will play out so well as AI surpasses human ability. The bottleneck might just move elsewhere while BIM/drafting lose value.
@oanaolt@rickwong888 I'm working on this with my current project Archipelago. Its a knowledge graph that allows types and rules to be defined within areas of concern. In theory it could address this class of problem but its still early.
@thdxr My project is all about how to work with the outputs of LLMs without always reading walls of text. Flow charts, knowledge graphs, interactive diagrams. Put the tools in their hands and let users choose their own adventure.
@nicbstme@heyitsyashu I've landed somewhere in between. Provide the MCP server with built in tools but then also the means to customize it on the go with a modular plugin architecture.