Calibra might help to put some benchmark numbers on your prototypes: https://calibra.swival.dev/

Just start building my man. Would highly advise keeping it minimal, just use OpenAI Agents SDK. Most of these frameworks are overengineered af

You can ask Claude or ChatGPT for help, or study the architecture of current well-known open-source agents.

I think the best way for that new thing is doing some real thing/join a team building an Agent, and get some know-how.

If you want something to actually build against mastra.ai is worth a look. its typescript framework and it covers the stuff you'll run into pretty quickly. workflows, memory, RAG, observability. I didn't build anything production ready with it but it's been great for getting a feel for how these pieces fit together

We wrote a couple of articles about it, but we were more focused on building on top of CLI-based agents like Claude Code (imagine a lead outreach agent based on Claude Code as an API).

21st.dev/community/blog

Text in, text out. It's all a bunch of prompts. Use the APIs directly and avoid as many abstraction layers as possible. Everybody is figuring it out all at the same time.

I find HN to be a bad resource to ask for learning resources. I previously asked for help in learning how claude works but no responses.

Maybe one pointer for others is that people are genuinely curious about learning new things, but as experts we choose not to engage these types of posts, why is that?

OP, in your case you need to move from theory into actually building your own agent and make it do things. Start by solving small problems and then make them more complex over time.

The best way to learn to build an agent is to learn and use pi.dev . The homepage is a masterclass of explaining the main loop with 4 tools (read write edit bash) and how they enable any flow. Skills and Agents.md on how the agent can be guided.

If you understand the above, you're 80% of the way to building any purpose agent. System prompt, project prompt, and the tool call loop.