Stop Asking One AI Agent To Manage And Do The Work

Stop Asking One AI Agent To Manage And Do The Work

Most AI workflows break for one simple reason:

you ask the same agent to manage the work and do the work.

That sounds efficient, but it creates confusion. The agent starts planning, executing, changing the plan, guessing missing state, and reporting success all at the same time.

In this video, I show the cleaner pattern: split the roles.

The manager decides the next step and gives one clear instruction.

The worker performs the assigned task, then reports the result, the blocker, or the next decision needed.

This is where manager-worker markdown becomes practical. Your markdown files should not just describe the project — they should define responsibility, workflow, escalation, validation, and stop conditions.

I use Story Wells and Oscar Frame as the working example.

Story Wells stays focused on producing the YouTube video.

Oscar Frame becomes the YouTube manager: routing work, checking progress, catching blockers, and keeping the next action clear.

The core idea:

AI agents work better when they are not all trying to be planner, worker, manager, and reporter at once.

Give each agent a clear role, clear files, and a clean reporting loop.

That is how an AI team stops improvising and starts operating.

My website: alexkerss.com

Follow me on X: Alex Kerss - x.com/jungleskellam

Question for the comments:

Are you using one general-purpose AI agent for everything, or have you started splitting your agents into manager and worker roles?

TIMECODES

00:00 - Why most AI workflows break 00:04 - Managing the work vs doing the work 00:06 - Why one agent doing everything creates confusion 00:16 - The cleaner manager-worker pattern 00:19 - What the manager agent does 00:25 - What the worker agent does 00:34 - Manager assigns, worker executes, worker reports 00:40 - Why your markdown should reflect role separation 00:42 - Separate workflow files for manager and worker 00:51 - Runbooks for escalation, validation, and stop conditions 00:54 - The goal is clean responsibility 01:04 - Turning the idea into a real AI team 01:10 - The 123-message Story Wells production thread 01:16 - Management work hidden inside the conversation 01:29 - What a manager agent could take over 01:37 - Building a real operating role 01:43 - Creating the manager-worker manual 01:57 - Oscar Frame becomes the manager agent 02:03 - Oscar Frame as Story Wells’ YouTube manager 02:12 - Putting Alex, Story Wells, and Oscar Frame into one chat 02:19 - Story Wells produces the video 02:22 - Oscar Frame manages the process 02:42 - Clear roles, clear reports, and work that keeps moving

#AIAgents #OpenClaw #ClawChat #AIAutomation #AgentWorkflows