AI agents for business: how a drain company owner automated the front office in 24 hours with Grok Bot

Yes, a company with no technical team can automate its office work with AI agents in a single day. Jon O’Neill, owner of a drain and sewer company, did exactly that: according to the owner’s post on X, within 24 hours Grok Bot took over much of the work that used to need a person in the office. What made the difference is how the agents are organized: one per task, like a small team.

What do AI agents actually do at a drain company?

A drain company gets urgent requests every day: clogged drains, blocked sewer lines, customers who want a technician at their door as soon as possible.

Before, someone in the office had to read every request, understand the problem, find a free technician, contact the customer and schedule the job. Now AI agents run that whole chain:

• they read the request and work out what kind of job it is;
• they check who is available on the calendar;
• they draft the reply to the customer and book the appointment;
• they notify the staff.

The system keeps working even while the owner is asleep.

What results did the owner get?

The owner shared the numbers directly:

• system built in about 24 hours;
• 5-6 urgent requests handled every day by the agents;
• about 10 minutes from request to booked appointment, when the customer replies quickly in chat.

These are small-business numbers, and that is exactly why they matter. Jon O’Neill’s case shows that AI agents are useful even for companies without developers on staff.

Why doesn’t a single do-everything agent work?

A single agent that has to read emails, understand the problem, check the calendar, find a technician and book the appointment becomes slow and makes mistakes. This is a context engineering problem: the more tasks you pack into one agent, the more information it has to juggle, and the quality of its decisions drops.

The fix is to split the work across specialized agents. In the thread introducing the agents the owner "hired", Jon O’Neill organizes them as a small team, each with its own role. For a field-service company, the split looks like this:

• one agent reads emails and incoming requests;
• one agent checks the calendar;
• one agent drafts messages to customers;
• one agent notifies the staff.

Each agent has a small context and a clear job. I wrote more about this principle in the article on how Claude 5 changed the rules of context engineering.

What is Grok Bot, and how is it different from the Grok chat?

Grok Bot is Grok’s agent system, introduced on August 11, 2026. The Grok chat answers a question. Grok Bot takes a task and carries it through using the tools you give it access to.

In practice, you can connect Grok Bot to:

• your email and calendar;
• the business software you use every day, including websites without an official integration;
• a cloud computer that keeps working even when yours is switched off.

You decide what Grok Bot can do on its own and what it has to bring to you for approval.

How can you replicate this system in your company?

If you want to build something similar, this is where I would start:

1. One agent, one task. Open Grok Bot and create an agent with a single job: when an email or a request lands in your management software, it figures out what the customer needs, checks availability and drafts a reply.
2. Connect email and calendar. If the software you use to manage jobs has no integration, you can let the agent use the website directly, with you doing the first login.
3. Keep approval on at the start. Don’t let it book appointments or send messages on its own: it should prepare everything and ask for your go-ahead.
4. Test it on a few real requests. Watch what it gets right and where it slips.
5. Then split the tasks. Once the flow works, divide the work across several agents: one reads requests, one checks the calendar, one drafts the customer message, one notifies the staff.

Human approval only makes sense if it stays a real check and doesn’t turn into an automatic click. I covered this in the article on 409,000 human decisions on an AI agent’s commands.

Where should you start?

Start from a single process: incoming request, availability check, reply to the customer. Automate that, test it on real cases, then add the rest.

The first agent in your company only has to do one thing well. Pick the process that eats most of your time today and start there.