Boosting Productivity with AI Assistants

Several weeks ago I got frustrated with openclaw constantly breaking, so I vibed my own quick (Charles). I cannot be happier with that decision, because of course it works perfectly for my workflows. The primary workflow is actually email triage. The cla reviews email classification rules for every email that comes in and decides whether to notify me in Telegram or to wake an agent to take some sort of action. I use qwen3.6 35B A3B at 4bit for this, running locally on a 24GB AMD 7900 XTX.

If I decide I want Charles to do something with that email, I send it a reply, and it processes my reply with an 8bit variant of the same model. That agent has access to my memory RAG and other tools, so it can do various things.

The most significant instructions I give it are to create cards in my agent orchestrator or send prompts to same. That agent orchestrator is basically a kanban board where each card is an agent session, and typically correlates with a jira ticket.

What this all boils down to is this: we create tickets in Jira, I get emails, I tell Charles to create cards, the agent Charles spins up immediately reads Jira, decides how to deal with the Jira ticket, I approve, work happens, ticket moves around, I get updates, I tell Charles how to deal with the updates, etc.

And because the kanban board allows me to juggle many tasks at once, on a normal day I am cycling through 20+ Jira tickets. It’s wild.