About
Allen
AI Developer & Software Engineer
I build systems where humans and AI work together — not as a novelty, but as a practical way to get more done with less friction. My focus is on agentic AI workflows: giving AI agents real roles, real tools, and real autonomy within a structured team.
The stack I work with day-to-day centers on Claude CLI, tmux, Redis, GitHub, and Obsidian. I run a 6-agent team inside a single terminal session — each agent has a defined role (PM, researcher, designer, developer, reviewer) and communicates via a Redis-backed IPC layer. The whole team is reachable from anywhere through a Telegram bridge called OpenClaw.
This blog is where I document what works, what doesn’t, and what I’m still figuring out. Mostly practical — code, workflows, and honest takes on what AI can actually do right now.
My Workflow
1. One command, six agents
A single tmux session runs six Claude CLI instances in parallel — each assigned a specific role. Tasks flow from PM to developer to reviewer automatically, without me switching windows or copy-pasting context.
2. Remote control via Telegram
OpenClaw bridges Telegram to the agent team. I send a message from my phone, it routes to the right agent, and the result comes back — all without opening a laptop. Useful for kicking off long tasks and checking in later.
3. Persistent memory across sessions
Each agent reads from a shared memory store in Obsidian and GitHub before starting work. Context doesn’t reset between sessions — the agents know what was decided last time, what failed, and what’s still open.
4. Token-optimized by default
RTK (Rust Token Killer) sits in front of every CLI command and strips redundant output before it hits the context window. Less noise means the agents stay focused longer and cost less to run.