The 4-Category System That Lets AI Handle My To-Do List Autonomously

Yesterday at 5:02am, while I was asleep, my @openclaw AI agent read through my 23 tasks for the day.

It proactively drafted a LinkedIn post, wrote two follow-up emails, and flagged three items that needed my attention.

I woke up two hours later to find half my to-do list already done.

It took me 2 hours to build this system from scratch.

Here's my step-by-step guide that you can copy.

PS: I made a custom GPT that interviews you (takes 3 minutes) then delivers a custom AI Agent Blueprint.

The blueprint contains the skills/data you need in order to launch an effective custom AI agent.

Grab it (free) here:

http://return-my-time.kit.com/4b0927a64c

----------------------------------------------

The Problem I Was Solving

Every morning I'd open Todoist and see 20+ tasks due today.

Some I could delegate. Some I had to do myself. Some needed more context before anyone could touch them.

Sorting through them took 30+ minutes.

Not the work itself, just figuring out what work needed to happen and who should do it.

That's exactly the kind of cognitive overhead AI should eliminate.

So I built a skill for my AI agent (@openclaw) that runs every morning at 5am. It pulls every task due today, categorizes each one, and drafts content for everything it can handle.

By the time I wake up, my to-do list is already triaged. Half the work is drafted and waiting for my review.

The 4-Category System

The skill sorts every task into exactly 4 categories:

✅ Can Complete Tasks the AI can fully handle.

Draft the email, write the LinkedIn post, create the meeting agenda, research the person I'm calling today. These get drafted immediately and flagged for my quick review.

🟡 Can Partially Help Tasks where AI can do 80% of the work but I need to execute the final step.

Example: "Email the accountant about tax documents." AI creates a checklist of every document needed, gathers the links, drafts the email body. I just hit send.

🔍 Need More Info Tasks that are missing context.

The AI flags them and asks me specific questions to fill the gaps. "Why are we doing this? What does 'done' look like? Who has access?"

❌ Requires You Physical tasks, financial decisions, anything that needs my direct judgment or presence.

The AI knows it can't handle these and leaves them alone.

Every single task gets categorized. Nothing falls through the cracks.

What This Actually Looks Like

Here's a real example from yesterday morning.

1. Task: "Draft LinkedIn post about the Clawdbot Masterclass article"

The AI saw this at 5am. Categorized it as "Can Complete." By 5:02am, it had:

• Read the full Masterclass article from my Drive
• Analyzed Corey's LinkedIn voice (pulled from my brand-voice references)
• Drafted a 200-word post with a hook, 3 key points, and a CTA
• Flagged it for my review in Discord

I woke up at 6:30am, read the draft, made one small edit, and scheduled it.

Total time: 3 minutes. The AI did the other 27 minutes of work overnight.

2. Task: "Follow up with Ricardo about Zoom test"

The AI saw this was a follow-up (I have follow-up workflows built in). It:

• Checked the task comments to see this was attempt #2
• Read the previous message I'd sent
• Drafted a shorter, different-angle follow-up
• Showed me the draft and asked if I wanted to send it

I approved it, sent it, and the AI logged the attempt in Todoist comments automatically.

I'm not asking AI "help me with this task." I'm reviewing work it already completed.

The Technical Build (For Anyone Who Wants to Replicate This)

I built this as a custom skill for Clawdbot (my self-hosted AI agent platform). But the concepts work in any agent harness.

The skill has 3 main components:

1. API Integration - Pulls today's tasks from Todoist using their API. I use a filter that matches my exact "Today" view (responsible for me OR unassigned, top-level tasks only, due exactly today).
2. Classification Logic - Each task gets analyzed against a decision tree:

• Does it have "Why/Done/Access" context in the description? (If no → "Need Info")
• Can it be completed digitally? (If no → "Requires You")
• Does it require my strategic judgment? (If yes → "Requires You")
• Can I draft the full deliverable? (If yes → "Can Complete", else "Can Partially Help")

3. Content Drafting Engine - For "Can Complete" and "Can Partially Help" tasks, the AI immediately drafts the output using my brand voice references, pulls context from linked docs, and presents the draft in Discord for quick review.

The skill runs via cron at 5am every day. I don't trigger it. I just wake up and check Discord for the triage summary.

Tools used:

• Clawdbot (AI agent platform, self-hosted)
• Todoist API (task management)
• Python scripts (API calls + JSON parsing) (@openclaw wrote this code itself)
• Brand voice references (markdown files with my writing style)

Time to build: ~2 hours to write the skill, test the API calls, and wire up the cron job.

Time saved per day: 30+ minutes of cognitive overhead, every single morning + variable time saved depending on "Can Complete" tasks it handles autonomously.

Why This Matters Beyond My To-Do List

The entire point of this article is shifting how you use AI.

Most people are stuck in the "AI as co-pilot" phase. You're flying the plane, and AI is helping with navigation.

That's fine for some work. But the real leverage is in "AI as autopilot" where you set the destination and walk away.

The system I built doesn't wait for me to ask.

It proactively pulls my tasks, makes decisions about what to handle, drafts the work, and reports back. I review, approve, and move on.

That's the behavior I want from an assistant. Not "help me when I ask." But "handle what you can, flag what you can't, and show me your work."

And here's what most people miss: this isn't that hard to build anymore. AI agents can now:

• Make API calls to any service with documentation
• Classify and prioritize based on clear rules
• Draft content in a specific voice (if you give them references)
• Run on a schedule without human intervention

The only remaining bottleneck is the mindset shift from "AI helps me work" to "AI does work for me."

What You Can Do Right Now

You don't need to build a custom skill to benefit from this approach. Start smaller:

1. Define your categories

What work can AI fully handle? What can it partially handle? What needs you? Write those buckets down explicitly.

2. Give AI more context

Most tasks fail because the AI doesn't have enough information. Add "Why/Done/Access" context to your tasks.

Why are we doing this? What does completion look like? Who has access to the systems/docs needed?

3. Batch your AI usage

Don't ask AI for help task-by-task.

Give it your full list once, have it triage everything, then review the output in one session. You'll work faster and catch patterns.

4. Build iteration into your workflow

My AI logs every follow-up attempt in Todoist comments.

When a task repeats, it reads the history and tries a different approach. Small features like that 10x the value.

5. If you can code (or know someone who can)

The entire Todoist triage skill is ~200 lines of Python + a markdown skill file. I have zero technical skill and zero coding knowledge.

Let your agent write the code it needs to get the job done.

Try This

Pull up your to-do list right now. Pick 5 tasks. For each one, ask yourself:

• Could AI draft this if it had the right context?
• What context is missing?
• If I gave AI that context, how much of this task could it handle?

Then give AI (OpenClaw, Claude Cowork, whatever you use) one of those tasks with full context and see what it does. Don't edit your ask. Just see what comes back.

You'll be surprised how much it handles. And once you see that, you'll start thinking differently about what "delegation" actually means.

Key Takeaways

• Stop treating AI like a search engine. Build systems where AI works autonomously and reports back.
• Triage is the highest-leverage automation. Sorting "what needs me vs what doesn't" saves more time than any single task.
• The 4-category system (Can Complete, Can Help, Need Info, Requires You) works for any to-do list, any industry.
• Context is everything. Tasks with "Why/Done/Access" descriptions get handled. Vague tasks get flagged.
• You don't need to be technical to start. Define your categories, batch your work, and iterate with AI before building automation.

-------------------------

Want to build something similar?

My Agent Blueprint Builder custom GPT (free) can help.

It'll give you a blueprint of the exact data/skills you need to create an effective custom AI agent.

Get it here:

http://return-my-time.kit.com/4b0927a64c