The OpenClaw Business Guide

Security Warning

OpenClaw is new, powerful software — and with power comes real risk. If you don't know what you're doing, getting hacked is not a hypothetical — it's likely. You are giving an AI agent access to your business systems, credentials, and data. A misconfiguration, an exposed API key, or a careless permission can compromise everything.

It's very early days. The ecosystem is evolving fast. Security tooling, sandboxing, and best practices are still catching up. Products that make this safer are weeks to months away.

Tread carefully. Don't rush into production with business-critical systems.
Hire people who know what they're doing. Server security, API management, and AI agent configuration are real skills. This is not a weekend project.
Consider waiting. If you're not technical, it may be worth waiting a few weeks or months for safer, more polished tools to emerge.
Never store credentials in plain text. Never expose your server to the open internet without hardening. Never skip the basics.

By Dan Peguine — a living document, last updated February 2026

This guide is written for you AND your agent. Paste it into your OpenClaw and it will start identifying workflows it can help with in your business.

How I Got Started with OpenClaw

I help run a specialty retail business with my parents — two physical stores, an e-commerce site, staff scheduling, inventory, monthly supplier orders from overseas.

Mapping the Business

The agent started asking questions:

"What happens every Monday?"
"Who handles the schedule?"
"What's your most annoying recurring task?"
"What breaks when someone goes on vacation?"

By the end, we had:

A full team directory (names, roles, which locations, contact info)
Weekly, monthly, and yearly business cycles mapped out
A prioritized list of pain points

The result: An agent that actually knows the business.

What You Need Before You Start

Access to your key people — even a quick call to whoever knows the workflows
Willingness to explain the dumb stuff — things you take for granted, the agent doesn't know

How to Find Workflows Your Agent Can Take Over

1. Record everything

The single best thing you can do: record yourself (or the person who does the work) walking through the annoying task. Screen recordings, voice notes, whatever captures it.

Screen record with Loom — walk through the workflow while narrating what you're doing and why. My dad recorded his most tedious workflow this way and the agent processed the video, extracted frames every 30 seconds, transcribed the narration, and correlated both to understand the full process.
Send voice notes — talk to your agent like a person. Describe the task, the edge cases, the things that annoy you. WhatsApp voice notes work great. Just dump everything — the agent takes notes and asks follow-up questions.
Screenshot your tools — take screenshots of the dashboards, spreadsheets, and tools you use. The agent can see them and start mapping what connects to what.

2. Let the agent interview you

Don't try to write a perfect spec. Instead, give your agent the role of a business analyst and let it ask questions. Tell it:

"I want you to understand how my business works. Interview me. Ask me about my daily workflows, what I do manually, what's repetitive, what's annoying. Take notes and organize them."

The agent will ask surprisingly good questions — things you wouldn't think to mention. It'll organize your messy brain dump into structured workflows with inputs, logic, outputs, and human checkpoints.

3. Dump everything, organize later

The mistake people make: trying to think of the "right" workflows to automate first. Just dump everything — the agent will help you prioritize:

Every task you did this week that felt repetitive
Every report you generate manually
Every follow-up you forgot to send
Every spreadsheet you copy-paste between
Every thing that makes you think "there has to be a better way"

Your agent will sort through it and prioritize.

4. Start with one workflow

Pick the one that's most annoying AND most repeatable. For us it was shift scheduling — happened every week, same process, same people, same tools.

The agent drafts the workflow, you review it, test it with human approval at every step. Gradually remove checkpoints as you trust it more.

Workflows That Work Well

Great for agents

Scheduling & coordination — collect availability, draft schedules, send reminders, manage calendars
Report generation — pull data from Shopify/GA4/social, compile into readable reports, deliver on schedule
Document processing — collect invoices, certificates, compliance docs, package them up
Monitoring & alerts — watch ad campaigns, flag broken tracking, check competitor prices, daily health checks
Email to action workflows — parse inbound emails, extract info, draft responses or proposals, route to the right person
Supplier ordering — process inventory data, fill order forms, track shipments
Reminder systems — recurring deadlines, license renewals, payment due dates

Still needs a human

Final approvals on anything customer-facing
Exception handling — when something doesn't fit the pattern
Strategic decisions — pricing, hiring, partnerships
Ambiguous communication — when context matters more than content
Financial transactions — agent reminds, human clicks "pay"

Our Specific Automations (So Far)

Weekly sales report: 2 hrs manual → 5 min review. Human: glance + comment.
Monthly order form: 2-3 days → 4 hrs. Human: verify + send.
Customs PDF combining: 1 hr per shipment → 2 min script. No human needed.
New product imports: 30 min/product → 5 min/product. Human: review + publish.

What Worked, What Broke, What's Still Human

What Worked Immediately

1. The Business Context Document

We created one file (BUSINESS.md) with everything: company overview, team, products, suppliers, logistics chain, seasonal patterns, key contacts. The agent references it constantly.

2. Memory System — these come with OpenClaw out of the box

Daily notes capture what happened. Long-term memory holds curated lessons. The agent wakes up fresh each session but reads these files first, so context persists.

3. Cron Reminders — these come with OpenClaw out of the box

"Remind me on the 14th of every month to pay the invoice." Done. Never forgotten again. The agent pings the right person at the right time.

4. API Integrations

Once the agent had Shopify API access, it could pull orders, check inventory, even create draft products. Same pattern works for any system with an API.

What Broke (And What We Learned)

1. Assumed Context

Problem: The agent didn't know aliases and jargon. Fix: Add aliases and local jargon to the business doc. Treat the agent like a new hire — explain everything.

2. Over-Eager Automation

Problem: Agent tried to send emails without asking. Fix: Explicit boundaries: "Anything external-facing needs human review before sending."

What's Still Human (And Should Stay That Way)

Purchasing decisions — Agent suggests, human approves.
Hiring and HR — Never automated. Full stop.
Customer escalations — Agent can draft, human sends. Tone matters.
Payments — Agent can remind and prepare, human clicks the button.
Anything public — Tweets, emails to customers, blog posts. Human reviews.

The Golden Rule: If it leaves the building, a human checks it first.

How to Share This With Your Agent

This guide is designed to work as a prompt. Copy the sections that are relevant to your business and paste them into a conversation with your OpenClaw agent. Try:

"Here's a guide on how to identify workflows you can help me with. Read it and then interview me about my business. I want you to find the workflows that are eating up my time."

Your agent will read the guide, understand the approach, and start asking you the right questions. That's the whole point — you don't need to figure out what to automate. Your agent will help you find it.

Pro tip: Send this guide link to your agent along with a Loom recording of your most annoying weekly task. That's all you need to get started.

This guide is updated regularly. Follow me on X (@danpeguine) for real-time updates, or check back here, or ask your agent to check this regularly and update you about any new changes.

Quick Start Checklist

Day 1

Give your agent a name and role
Spend 30 min explaining your business
Identify your #1 pain point workflow
Record a Loom of someone doing that workflow
Establish boundaries (what needs approval)

Week 1

Let the agent interview you about all your workflows
Document one workflow in detail
Set up one recurring reminder or report

Month 1

Automate 2-3 recurring tasks
Set up the memory system
Measure time saved vs. before
Train your team on how to talk to the agent

This guide is based on real experience onboarding an AI agent for a specialty retail business. Last updated: February 4, 2026.

Follow me on X (@danpeguine) for real-time updates.