Now

Next up: structuring personal/project knowledge as a graph rather than flat notes, to give agents better long-term context.

Building automated review loops where one agent's output is checked and critiqued by another before it reaches me, to catch mistakes earlier and raise the bar on AI-generated code.

Exploring how to feed coherent, up-to-date repo context to coding agents (memory, CLAUDE.md-style instructions, retrieval) so they stop re-deriving things I already know.

Designing the surrounding harness — tools, guardrails, permissions — that determines whether an autonomous coding agent is useful or dangerous.

Explored spec-driven development mainly using Google's spec-kit framework to keep AI-assisted development aligned with actual intent.

Compared CLI agentic dev environments using lower class models (Kimi K2, Llama, Qwen, etc.) to find more cost-efficient ways for LLM-supported development.

Read the CISSP study book to improve my overall knowledge regarding IT security (from a defensive perspective).

Built a full agent setup on Openclaw to assist with my own development and build workflows.

Ran local LLMs via Ollama to evaluate self-hosted models as an alternative to cloud APIs.