What happens when a ticket lands on it
Think of it as a junior technician who is very fast at the first twenty minutes of a ticket and honest about where its knowledge runs out. The sequence is the same one your own people follow.
- It finds the machine. A ticket is tied to a device once, by you, in the mapping screen. A ticket with no mapping is left alone rather than guessed at.
- It looks before it touches. It runs read-only checks on the real endpoint - services, disks, event logs, network state - and reads what comes back. The next command depends on the last answer, not on a fixed script.
- It proposes a fix. Anything that changes the machine is shown to a person first, in plain words, with the exact command underneath.
- It checks the fix held. After a change it re-tests the original complaint with a separate command. A command that exited cleanly is not treated as proof that the problem is gone.
- It writes the note. The summary goes to the ticket as an internal note by default, so nothing reaches your customer unreviewed.
What it will not do
The useful property of a technician you can trust is that they say “I cannot tell” when that is the truth. The product has a dedicated outcome for it.
- It can decline to conclude. When the evidence does not support a diagnosis, the session ends as “insufficient evidence”, with what it found, what it could not confirm and a specific next step for a human. That is recorded as its own result, not counted as a success.
- It does not loop. After a fixed number of failed fix attempts on one problem it stops proposing fixes and says so. Your administrators can raise or lower that limit.
- It does not self-approve identity changes. Deleting a user, deleting a group or resetting a password in Microsoft 365 always needs a person, even on a machine you have set to run everything automatically.
- It does not take a third party's word for the cause. A session that is marked resolved with no command run behind it is flagged for review on our side.
Where a person stays in charge
| Kind of step | Who decides |
|---|---|
| Read-only checks (list services, read a log) | Your per-machine setting: run automatically, or wait for a click |
| Anything that changes the machine | A person approves the exact command |
| Disk wipes, shadow-copy deletion, disabling Defender, boot configuration, user deletion | Blocked outright; you can add your own never-allowed list |
| Microsoft 365 identity actions | Always a person |
The default for a newly added machine is the cautious one: every command waits for a click. You loosen it per machine when you have seen enough to trust it.
Honest limits
An AI technician can only act where an agent is installed and connected. It cannot swap a failing drive, and it cannot decide a business question for you. Some jobs also need a person at the machine: a major macOS upgrade, for example, needs a signed-in user with the right privileges, and the product says so instead of pretending it can finish remotely.
If a problem keeps coming back, the cause is rarely something one more command will find. That is the moment it is built to hand over, with its evidence attached, rather than keep trying.
Teaching it your fixes
When a person solves something worth repeating, the steps can be saved as an Autopilot: written instructions that a human reads once and saves. A new Autopilot starts at the most cautious level and earns more freedom through verified runs. The AI still reasons for itself about anything no Autopilot covers.
AI technician, chatbot or script: which is which
The phrase gets applied to three different things. A chatbot talks to the end user and never touches the device. A script does one fixed thing when triggered. An AI technician, in the sense used on this page, does neither: it works on the machine, choosing each next check from the last result, with a person holding the keys to anything that changes it.
If you are comparing vendors, ask which of the three you are actually being sold. The comparison of AI support and scripted automation lays out how each fails, and endpoint remediation for MSPs covers the device-side detail.
Frequently asked questions
What is an AI technician for MSPs?
It is software that picks up an endpoint ticket and investigates the real machine the way a junior technician would, then proposes a fix and checks that it worked. Changes to the machine wait for a technician's approval unless an administrator has allowed otherwise.
Does an AI technician replace my technicians?
It takes the first part of a ticket, the investigation, and hands back what it cannot settle. Your people still approve changes, own the never-allowed list and decide what happens when the evidence is thin.
Which PSA does it work with?
Autotask is the supported PSA today. A ticket that has not been mapped to a machine is left alone rather than guessed at, and the summary is posted back as an internal note by default.
How is an AI technician priced?
The price is published on the pricing page: a base fee, a per-endpoint rate that tapers, and a charge for each ticket that is actually resolved. Technicians are unlimited.