← all notes
AI automation claude Note 003 · March 15, 2026 · 9 min · guide · Updated November 14, 2026

A Non-Engineer's Guide to Building Autonomous AI Agents with Claude Cowork

A non-engineer's guide to building autonomous AI agents with Claude Cowork. Daily Slack messages, research reports, and file organization on autopilot.

the whole walkthrough, if you'd rather watch than read.

Updated November 14, 2026: Scheduling has changed since this guide was published. The sections on execution location and keeping a computer awake have been revised using Anthropic’s current scheduling documentation. The walkthroughs below reflect the original interface and may differ from your current app.

You keep hearing about “AI agents,” autonomous systems that do real work without you hovering over them. But every guide you find is written for developers. Full of API keys, cron jobs, and Docker containers.

I’ve been looking for a way to change that, a way for anyone to build their own agents that wake up on a schedule, complete a task, and go back to sleep. No code required.

That tool finally exists. It’s called Claude Cowork, and it’s part of the Claude Desktop app. If you’ve heard of Claude Code (Anthropic’s coding agent for developers), Cowork is built on the same architecture but wrapped in an interface anyone can use. No terminal. No debugging. No code to write or maintain.

What can Cowork do? You describe a task once, set a schedule, and Claude handles it going forward: posting Slack messages, running research, organizing files, building reports. Automatically. On repeat.

In this guide you’ll learn: what autonomous agents are, how to set one up, three ready-to-use examples, and what to do if you want your agents to run even when your computer is off.


What Is an Autonomous Agent?

In the simplest terms: an autonomous agent is an AI that does work for you on a schedule, without you having to ask every time. You describe a task once, set a schedule, and the AI handles it going forward.

Think of it like this: you’re hiring a tireless assistant. You give them a job description once, and they show up and do the work every morning, every week, or whatever cadence you choose. They never forget, never get tired, and never need a coffee break.

What you can automate

  • Daily morning briefings: Claude searches the web, gathers news in your field, and drops a summary into your calendar or files
  • Slack messages: daily standup prompts, weekly metrics summaries, or team digests posted automatically to a channel
  • Research reports: monitor topics, competitors, or markets and receive formatted findings on a schedule
  • File organization: Claude cleans up your Downloads folder, renames files, and sorts them into the right places
  • Spreadsheet updates: refresh data, recalculate formulas, and format weekly reports
  • Presentation generation: Claude creates slide decks from your data on a recurring basis

Getting Started: What You Need

Before diving in, here’s what you’ll need:

  • An eligible paid Claude plan: Check the current scheduling documentation for plan availability and organization settings.
  • Claude Desktop app: installed and fully updated (Mac or Windows). If you don’t have it yet, head to claude.com to download it.
  • Connectors: set up for any apps you want Claude to interact with (Slack, Google Calendar, Gmail, Notion, etc.)

To install connectors: Open Claude Desktop → Click the puzzle piece icon (Customize) → Connectors → Browse Connectors → Find the app you need → Click Connect and authorize.


Usage and Cost Considerations

Start with one representative task and review its actual usage before increasing frequency. Tool use, context, and task complexity make a fixed number of tasks per plan an unreliable sizing rule. The earlier task-count recommendations have been removed because they were not measured capacity limits.

Where scheduled tasks run

Current scheduled tasks can run remotely while your computer is asleep or the desktop app is closed. Tasks that need local files or apps run locally. On Pro, Max, and Ultra, older local tasks created before October 6 continue to run on the computer. Confirm your task’s execution mode before relying on it.

See Anthropic’s scheduling guide for the current setup steps and exceptions. Review run history and test the output before depending on a schedule.


Example 1: Daily Slack Message Agent

This agent posts a message to a Slack channel every day at 9:00 AM. It could be a daily standup prompt, a motivational message, or a summary of yesterday’s metrics. The result looks like any other Slack message, and your team might not even realize it’s automated.

Step-by-step

  1. Connect Slack. Go to Customize → Connectors → find Slack → Click Connect → Authorize access to your workspace.
  2. Open Cowork. Click the “Cowork” tab at the top of Claude Desktop to switch from Chat to Tasks mode.
  3. Start a new task. Click “+ New task” in the upper left corner.
  4. Type /schedule. This launches the scheduling skill. Claude walks you through a few questions.
  5. Write your prompt. Be specific. Here’s an example:

Every weekday at 9:00 AM, post a message in the #daily-standup Slack channel. The message should include: a greeting, a reminder of our team’s top 3 priorities this quarter, and a prompt asking each team member to share what they’re working on today. Keep the tone friendly and professional.

  1. Set the schedule. Choose your frequency (daily, weekdays, weekly) and the time (9:00 AM).
  2. Click Save. Done! Your agent is created.

What success looks like

After the first run, you’ll see a message appear in your Slack channel just like any other post. Claude will also rewrite your prompt based on what it learned, figuring out the exact channel ID, formatting, and connectors to use. Future runs will be faster and more reliable.


Example 2: Daily Research Agent

This agent searches the web every morning for news in your field and delivers a summary. You’ll wake up to a neatly formatted file on your Desktop with the top stories of the day, curated by AI to match your exact interests.

Step-by-step

  1. Connect Google Calendar (optional, if you want findings delivered as a calendar event) or any other app where you want the output.
  2. Open Cowork → + New task → type /schedule.
  3. Write your research prompt:

Every weekday at 8:00 AM, search for the latest news about artificial intelligence, specifically about AI agents, AI automation, and new product launches. Focus only on factual, verified news. Skip opinion pieces and speculation. Summarize the top 5 stories in a bullet-point format. Save the summary as a file on my Desktop called AI-News-[today’s date].md.

  1. Set schedule to Daily, 8:00 AM.
  2. Click Save.

What success looks like

You’ll find a new Markdown file on your Desktop each morning with a clean, scannable summary of the day’s top stories. Each entry will include a headline, a brief summary, and a source link.

Variations you can try

  • Competitor monitoring: “Search for news about [competitor names] and summarize any product updates, funding rounds, or leadership changes.”
  • Industry digest: “Search for developments in renewable energy policy and create a one-page briefing document.”
  • Job market tracker: “Search for new senior product manager roles at Series B+ startups and list them with company, title, and link.”

Example 3: Weekly File Organizer

This agent tidies up your Downloads folder every Friday afternoon. No more chaos: PDFs go to one folder, images to another, spreadsheets to their own spot. You come back Monday to a clean slate.

Step-by-step

  1. Open Cowork → + New task → type /schedule.
  2. Write your prompt:

Every Friday at 4:00 PM, organize my Downloads folder. Move PDFs to Documents/PDFs, images to Pictures/Downloads, spreadsheets to Documents/Spreadsheets, and presentations to Documents/Presentations. If a folder doesn’t exist, create it. Leave anything you’re unsure about in Downloads. When done, create a short summary of what was moved.

  1. Set schedule to Weekly, Friday, 4:00 PM.
  2. Click Save.

What success looks like

When the task completes, you’ll find a summary file listing everything that was moved and where it went. Your Downloads folder will be clean, and your documents will be neatly sorted into their respective folders.


Managing Your Agents

All your scheduled tasks live in one place. Click “Scheduled” in the Cowork left sidebar to see everything at a glance.

Action How to Do It
Pause a task Click the task in Scheduled → toggle it off
Run on demand Click the task → click “Run now” (great for testing)
Delete a task Click the task → click Delete
View history Click the task to see past runs, including any that were skipped
Edit instructions Click the task → edit the prompt or schedule

Tips for Writing Great Agent Prompts

Clear instructions help, alongside suitable tools, accessible data, evaluation, and review. Here’s what matters most:

1. Be extremely specific

Don’t say “Post something to Slack.” Say “Post a message in #marketing-updates summarizing our top 3 campaign metrics from this week. Include click-through rate, conversion rate, and spend. Format as a bulleted list.”

2. Tell Claude where to find and put things

Mention specific folder paths, channel names, calendar names, and file locations. The more explicit, the better.

3. Set the tone

If the output goes to other people (Slack, email), describe the voice you want: formal, casual, friendly, brief, detailed.

4. Test before you schedule

Run the task once manually (using “Run now”) before putting it on autopilot. Review the output. Tweak the prompt. Run again. Only schedule it once you’re happy with the result.

5. Use context files for consistency

Create small text files in your working folder (like about-me.md or voice-and-style.md) that describe who you are, what you do, and how you like things written. Explicitly include or reference the relevant files and verify that the task can access them.


Running while your computer is off

Use remote scheduling when the task can work with supported connectors and account files. Local tasks still depend on the computer and its apps being available; use supported power settings and review missed or failed runs.

The earlier hardware mouse-jiggler recommendation and always-on computer buying advice have been removed. They do not address application failures, expired access, or whether a scheduled result is correct.


Wrapping Up

Building your own autonomous AI agent used to require developer skills. With Claude Cowork, it doesn’t anymore. You describe a task in plain English, set a schedule, and let Claude handle the rest.

Start with one simple agent, maybe a daily research summary or a weekly file cleanup, and see how it feels. Once you trust it, add more. Before you know it, you’ll have a small team of AI assistants handling the repetitive work that used to eat up your mornings.

The barrier to building agents isn’t technical anymore. It’s just knowing that you can.

I build AI products and help teams put AI to work. These are my notes from the work itself.
Get Landon Notes by email

Keep going

these connect to this one
Development · 2025.06.20 The Strategic Software Decision: Code, Train, or Prompt? the decision I get asked most Career · 2024.09.15 How to Use AI to Prep for Your Next Job Interview 💼 people actually got jobs AI · 2024.10.14 LLM-assisted porting: lessons from Swarm-rb a revised engineering note