An AI Agent for $20 a Month. Does IT Know? 5 Rules Before Anyone Turns One On
The video uses a digital avatar of Erick Grau, CEO of Chibitek.
For years, "shadow IT" meant someone on your team signing up for a file sharing app with a company card. This week it started to mean something bigger: an AI agent that keeps working when nobody is watching, on a plan that costs about as much as lunch.
If you run a business or an MSP, the question is not whether this is useful. It probably is. The question is whether anyone in IT knows it's running.
What Changed This Week
OpenAI makes an always-on agent called Dots. Until now, it was limited to higher-priced ChatGPT plans. This week, OpenAI started testing Dots with a small number of users on ChatGPT Plus, the $20 a month plan, according to The Verge and Engadget. OpenAI's own help center says Dots are rolling out gradually to select Plus users.

Dots are also coming to ChatGPT Business Standard. The Verge and Engadget report the rollout covers Business Standard users, and OpenAI's help center describes it as rolling out to eligible Business Standard members in the United States.
Here is what a Dot can do, according to OpenAI and Engadget:
It is always on. OpenAI says Dots "can work towards your goals 24/7," and Engadget describes them as agents that can work on tasks "without any supervision or assistance."
It connects to a lot of apps. Through OpenAI's plugin ecosystem, Dots can "connect to over 4,000 apps."
It has its own cloud computer, and with permission it can also connect to and work on a person's own computer.
None of that is a scandal. It is a capable tool. But notice who makes the decisions: the person who sets it up chooses which apps to connect and whether to connect their computer. That person may not be in IT.
How Microsoft Built Its Version
Microsoft announced its own always-on agent at Ignite on November 17. It is called Autopilot, and Microsoft says it is beginning to roll out to customers in its Frontier early access program. Redmond Magazine and Ars Technica covered the launch as well.
Like Dots, Autopilot works without waiting to be asked. Microsoft says it can "watch Teams channels, follow up on emails, run recurring work, and manage a project," and The Verge described the same kinds of tasks.
What makes Autopilot worth studying is how Microsoft wrapped it for IT. In Microsoft's words:
"Each Autopilot has its own Microsoft Entra Agent ID and a named manager."
"IT can scope its access and block it anytime."
"Its sign-ins and actions are logged under its own name, so security teams have a clear audit trail."
Ars Technica's coverage reports the same controls: a unique ID for each agent, administrators who can restrict permissions or block it entirely, and activity logged for human review. ZDNet's French edition quotes the same identity, manager and logging details.

Read that list again. That is the checklist. You don't have to buy Autopilot to use it. Any agent your team turns on, from any vendor, should meet the same bar.
The two worlds are also starting to meet. OpenAI says it is working with Microsoft to bring its organizational "specialist dots" under Microsoft's Agent 365 governance controls, and Microsoft lists OpenAI among the new platforms connected to Agent 365.
5 Rules Before Anyone on Your Team Turns One On
These five rules are Chibitek guidance, modeled on the controls Microsoft built into Autopilot.

1. Give Every Agent Its Own Identity
Never let an agent run under a person's login. If it signs in as someone in accounting, every email it sends and every file it touches looks like that person did it. When something goes wrong, you can't tell the person from the agent. Our Agent Reach review asked the same question from a different angle: whose login is your AI agent using?
2. Give It a Named Owner
One human answers for what the agent does. Not "the marketing team." One person. Microsoft makes this part of the design with a named manager for every Autopilot. If an agent has no owner, it has no one to call when it misbehaves, and no one to turn it off when that person leaves the company.
3. Scope Its Access
Give it only the apps and folders the job needs. An agent that can reach thousands of apps does not need all of them to draft your weekly report. Start small and add access when there is a reason.
4. Log Everything It Does, Under Its Own Name
If the agent has its own identity, its sign-ins and actions can be logged as its own. That is what turns "something weird happened in the shared drive" into a clear answer about who did what and when.
5. Know Where the Off Switch Is, and Test It
Microsoft says IT can block Autopilot at any time. For every agent in your business, someone should know how to stop it, and should have actually tried it once. An off switch nobody has tested is a guess.
What This Means for MSPs
If you manage IT for other businesses, your clients' employees are going to find these agents before you do. The $20 price point means approval may never cross anyone's desk. A short, friendly policy works better than a ban: tell people what an agent needs before it touches company data, and give them a quick way to ask.
That is also the conversation we want to have with business owners. Our earlier post, Your Team Is Already Using AI. Do You Know How?, covers the people side of that conversation.
The Bottom Line
Always-on agents are now within reach of almost anyone with a credit card. Microsoft's design for Autopilot shows what good looks like: its own identity, a named manager, access IT can scope and block, and logs under its own name. Hold every agent to that bar, no matter who makes it.
Key Points
OpenAI is testing always-on Dots agents with a small number of ChatGPT Plus users on the $20 a month plan, and rolling them out to Business Standard.
Dots can connect to over 4,000 apps and, with permission, work on a person's own computer.
Microsoft's Autopilot gives each agent its own Entra Agent ID, a named manager, access IT can scope and block, and logs under its own name.
Five rules for any agent: its own identity, a named owner, scoped access, logs under its own name, and a tested off switch.
Frequently Asked Questions
Should we ban always-on agents? That's your call, but a ban can push them out of sight. Clear rules that every agent has to meet are easier to enforce and easier to live with.
Do we need Microsoft Autopilot to follow these rules? No. Autopilot is a useful example because Microsoft built those controls in. The same five rules apply to any agent from any vendor.
Is a $20 agent really a business risk? The price isn't the risk. The risk is an agent with broad access that nobody in IT knows about, running under someone's personal login.
What is the first thing to check? Ask who on your team has already turned an agent on, and whose login it uses.
Let's Find Out What's Already Running
At Chibitek, we help businesses keep technology effortless, with no geek-speak. If you're not sure whether anyone on your team has already turned on an AI agent, or what it can reach, contact us at hello@chibitek.com and we can walk through the five rules together.
Identity. Owner. Scope. Logs. Off switch.
We make IT effortless so you can create, disrupt, and grow!
Follow Chibitek on YouTube at @chibitekmsp, and follow Chibitek on LinkedIn.
Sources
All pages were reopened on Nov. 20, 2026, and every quote above was found word for word.
The Verge, "OpenAI is starting to bring Dots to Plus tier subscribers," November 20, 2026
Engadget, "OpenAI is testing Dots personal agents on its $20 per month Plus tier," November 20, 2026
Ars Technica, "Copilot is 'the new OS for work,' and now Office runs inside of it," November 2026
The Verge, "Microsoft thinks its new Copilot 'super app' will be as influential as Office"
Microsoft Tech Community, "What's new in Microsoft Agent 365: Ignite 2026," November 17, 2026
Chibitek, "An AI Agent for $20 a Month. Does IT Know?" (YouTube Short)








Comments