Our footer says our agents work the night shift. It was a joke when we wrote it. Since August it has become a product category.
In the space of eight weeks, the three largest consumer AI companies each shipped the same idea: an agent that does not wait for your next prompt. It has a computer of its own in the cloud, signs into the tools you already use, keeps working when you close your laptop, and comes back when it needs a decision.
What actually launched
- Grok Bot (xAI), 11 August. "Your team of always-on agents." Each bot gets a persistent cloud computer with a browser, files and a terminal, and can work on sites that have no clean API. It is in beta for SuperGrok and Cursor subscribers; enterprises are on a waitlist.
- Muse (Meta), 8 September, and Muse for Small Business, 29 September. A personal agent first, then a version that connects to Facebook and Instagram business accounts, Shopify, QuickBooks, Slack, Canva, Stripe, Notion and others. Meta states that Muse will not publish, send messages or spend money without the owner's approval.
- Dots (OpenAI), 29 September at DevDay. Always-on agents powered by GPT-6 Astra, each with its own cloud computer, reachable in ChatGPT, Slack and Teams, connected to more than 4,000 apps through plugins. When idle, a Dot does "proactive research" with read-only access. It started with the Pro tier.
Three companies, three names, one design: a colleague, not a chat window.
Why this matters more for a 12-person company than for a bank
A large enterprise has a team for each function. An SME has a founder who does sales on Sunday, invoices on Monday and supplier follow-ups at night. OpenAI's own research this year described AI as becoming the "first hire" for small businesses, used for estimates, product descriptions and customer replies rather than for building technology.
Always-on agents push that further. The jobs they are being demonstrated on are exactly SME jobs: a sales bot updating the CRM from call notes, an ops bot processing invoices from an inbox, an agent noticing a client was never billed and preparing the invoice for approval.
In Saudi Arabia the timing is not an accident either. 2026 is the Year of Artificial Intelligence, and Monsha'at and SDAIA ran an AI Week for SME owners in August across Riyadh, Jeddah, Al Khobar and Madinah. The appetite is there. The question is what to hand over.
What I would hand to an agent today
- Follow-ups. Quotes sent but not answered, invoices due, suppliers who promised a date. Low risk, high value, easy to check.
- First drafts. Proposal sections, product listings in Arabic and English, a weekly summary of what happened in the shared inbox.
- Watching. A competitor's price page, a tender portal, a regulator's announcements. Read-only, with a short note when something changes.
And what I would not hand over yet: paying anyone, signing anything, and answering a customer complaint in Arabic without a person reading it first. Meta's approval rule for Muse is the right default for everyone.
The part nobody puts in the launch video
Each of these products asks for the same thing: logins to everything. An agent with its own computer and your Shopify, Gmail and bank-adjacent tools is a new employee with no contract, no onboarding and no exit interview.
For a Saudi company, three questions come before the subscription:
- Where does the data go? PDPL applies to your customers' data whether a person or a bot reads it. Know which region the agent's computer runs in and what it keeps.
- Who is accountable? SDAIA's position, stated plainly this September, is that an AI system does not carry legal responsibility; the developer and the operator do. If your bot sends it, you sent it.
- Can you switch it off? Write down how. Revoke the tokens, not just the subscription.
The open-source route
Not every business should rent its night shift from Silicon Valley. A wave of open-source projects now copies the same pattern and runs it on infrastructure you control:
- Memoh: bots in isolated containers with their own memory, reachable from Telegram, Slack, email and a web interface.
- nanobot (HKU Data Science): a small, popular Python framework for a self-hosted personal agent with tools, memory and MCP.
- OpenMausBot and Codync: projects that explicitly aim to be open alternatives to Grok Bot and Muse, the second running coding agents as named bots on your own machine.
- Daemora and Nexora: self-hosted agent platforms with scheduling, multiple model providers and multi-agent teams.
The trade is the usual one. You keep the data and choose the model, including Arabic-first or locally hosted ones, but you own the maintenance, the security patches and the bad night when a bot loops. For many SMEs the honest answer is a hosted agent for low-risk work, and a small self-hosted agent for anything that touches customer data.
Where this goes
My bet: within two years, "how many bots do you run" will be as normal a question for an SME as "which accounting system do you use". The winners will not be the companies with the most agents. They will be the ones that wrote down, in one page, what each agent is allowed to do, who approves it, and how to stop it.
That page is the real product. The bot is the easy part.