Claude Code may end up becoming the personal agentic framework that I've been looking for.
In the past when I needed some agentic behaviour (example: answer questions based on some documents), I used to look at hand rolling a solution. It's time consuming and not particularly easy to build and deploy.
Lately I've just been using Claude Code to do the same. Connect it to relevant knowledge bases via CLI or MCP. Then create a custom Claude Code command to get it to answer questions using the right sources and it does a fabulous job without any of the headaches of chunking, embedding and vector databases.
That's not to say that RAG is dead. When you want to answer diverse questions over massive databases and token cost is an issue, then it is still the better architecture. For personal productivity though, I can get most of the automation I want by just customising Claude Code and using that as my agentic assistant.
Recent Articles
- • Running AI Models Locally on Consumer Hardware Jun 15, 2026
- • AI Coding for Free Using Local LLMs at Chennaipy Meetup Jun 08, 2026
- • Pi and Py: Demoing an Open-Source Agentic Coding Assistant Jun 05, 2026
- • Is Clean Code Still Relevant in the AI Age? May 04, 2026
- • AI Agents Need Hard Boundaries Apr 24, 2026