Give Your Support Inbox to an AI That Never Logs Off

Your inbox doesn't sleep. As of eleven days ago, neither does the thing that answers it.

The inbox that never closes

Dana runs a six-person home goods brand. Her mugs and linens ship to customers in a dozen countries, which sounded charming until she realized what it does to a support inbox. There is no closing time. There is no time zone where everyone is asleep at once.

Support requests land in one address: help@. By the time Dana opens it at 9 a.m., a dozen messages are already waiting. Where is my order. Can I return this. Does this fit a queen duvet.

She used to answer them all herself. Then her ops lead did. Then nobody had time, the queue backed up, and two slow replies turned into refunds she didn't need to give.

This is not a Dana problem. This is every small business with an inbox and a time zone problem.

You are the bottleneck, not your product

Here is how it works right now, in almost every small business:

• A customer emails a question.
• The email sits until someone opens the inbox.
• Someone answers, well or badly, depending on how the day is going.
• Nobody records what was said, so the next similar question starts from zero.

None of that is a product problem. It is an attention problem. You are the bottleneck. Your product is fine at 2 a.m. You are not awake to say so.

What is Grok Bot, and how is it different from a chatbot?

On August 11, xAI (now folded into SpaceX as SpaceXAI) shipped something it calls Grok Bot. Not a chatbot. A team of what the company calls always-on agents, each one with its own cloud computer, that sign into the tools you already use and keep working after you close your laptop.

A chatbot answers the question you type into it, right now, in that window, and forgets you the moment you close the tab. An agent like Grok Bot gets a job description instead of a single prompt. You tell it once: watch this inbox, draft replies, ask before you send. Then it goes and does that job, continuously, coming back only when it needs your approval or has finished.

That is the whole shift in one sentence: a chatbot answers, an agent finishes the job.

You do not need Grok Bot specifically to get this benefit. Nous Research's open-source Hermes Agent added a similar feature called Bot Mode within days of the Grok launch, letting you spin up named agent profiles, each with a job description and a profile picture, that you switch between like separate hires. OpenClaw does a version of the same thing if you are comfortable running it yourself. The pattern, not the brand, is what matters: a persistent worker with one job, not a chat window you have to keep prompting.

Isn't this just going to send weird emails to my customers?

This is the question every owner asks first, and it is the right one. The honest answer: it can, if you skip the step everyone is tempted to skip.

The mistake is turning an agent loose on full autopilot on day one. Don't do that. Every serious build of this starts the same way: the agent drafts a reply, and a human approves it before it goes anywhere. You watch the tone. You watch whether it invents a refund policy you never had. You correct it in plain language, the way you'd correct a new hire.

Only once a week or two of drafts read the way you'd write them yourself do you flip the setting that lets it send without waiting. That is not a technicality. It is the difference between an agent that helps your business and one that embarrasses it in public.

Salesforce's own research on conversational AI support puts a number on the upside once you get there: platforms like this can automate up to 80% of routine customer inquiries, the ones asking about order status, hours, or return policy, freeing your actual humans for the messy 20% that needs a person.

How do you actually set this up? A six-step build

Chris Ashby, who builds and ships small software products for a living, filmed himself doing this in nineteen minutes for his own support inbox. Here is the build, stripped down to the steps that transfer to any business:

1. Point it at the inbox that actually matters. Not your personal email. The address customers already use: support@, help@, or the shared inbox your team checks. Connect it through the Gmail or Google Workspace connector and authorize the one account.
2. Create one agent with one job. Give it a name and a plain-English brief: watch this address, draft replies for anything that lands, wait for approval before sending anything. Resist the urge to give it five jobs on day one.
3. Hand it a knowledge base, not a lecture. You do not need a perfect wiki. A single markdown file covering pricing, refund policy, common troubleshooting steps, and your tone is enough to start. If you don't have one written down, tools like Claude Code or ChatGPT can read your website or your docs and draft that file for you in minutes.
4. Set the routine and the cadence. Most setups check the inbox every fifteen minutes during business hours to start, then widen to 24/7 once you trust the drafts. Every unhandled thread it sees gets tracked, so it never answers the same customer twice.
5. Send it one real question. Not a test message, an actual customer question, one you know the right answer to. Read the draft it produces before you approve anything. This is where you catch a wrong refund policy or an off tone before a customer ever sees it.
6. Loosen the leash on purpose. After a stretch of clean drafts, you flip the setting from "draft for approval" to "send automatically, only ask me when unsure." That one setting is the entire difference between a tool you babysit and a teammate that works while you sleep.

One small but real data point on how far people are already pushing this pattern: a plumbing company owner posted on X this week that he's using a similar always-on agent to generate real customer calls, and estimated it is saving him roughly $2,000 a month versus what he was paying a marketing agency. That is one owner's anecdote from one thread, not an industry benchmark, but it is a useful signal that this is already leaving the demo stage and landing in ordinary service businesses.

Grok Bot vs Hermes Agent vs OpenClaw: which one should a small business start with?

All three do a version of "persistent agent with its own environment." They differ mostly in how much setup pain you're willing to trade for how much control you want.

That ranking comes from two independent hands-on builders (one running a livestreamed comparison, one filming a step-by-step build) who each tested all three in the days right after launch, not from a vendor's own marketing. If you have never touched an AI agent before, start with whichever one has the fewest steps between you and a working support inbox. You can always move to a self-hosted option later once you know exactly what you want the agent to do.

What this looks like in Dana's shop

A week in, Dana's agent is drafting every reply, and she is reading every one before it goes out. She is editing maybe one in five, mostly tone, sometimes a fact it got slightly wrong from an outdated page on her site.

A month in, she has stopped reading most of them. The routine handles order status and returns entirely on its own now. She only gets pinged for the handful of messages that smell like a real complaint or a question the knowledge base doesn't cover.

Six months in, the support inbox is no longer a thing she thinks about before bed. Her actual humans spend their time on the customers who need a person: the ones with a broken order, a custom request, a reason to become a repeat buyer. Same team. Same inbox. A completely different relationship with the hours she used to lose to it.

That loop, drafts in, approval out, exceptions to a human, is the one worth building first no matter which tool you pick. Everything else is a setting you flip once you trust it.