Completing sign-in…

Back to the blog
MSP leadership·Nov 12, 2026·9 min read

AI Agents for MSPs: What They Do, How They Work and How to Choose One

AI agents for MSPs read tickets, act through your PSA, RMM and Microsoft 365 tooling, verify the result and document it, escalating what they cannot finish. This guide explains how they differ from copilots and scripts, which work they handle today, what your stack needs, how vendors price them, and the questions to ask before you buy.

MvMartijn van der SchaafResearch & editorial, Aegentics
AI Agents for MSPs

AI agents for MSPs are software workers that take tickets and recurring tasks off the service desk: they read the request, gather context from the PSA, RMM and documentation, take the action, check the result and record what they did. This guide explains what they are, which work they handle today, what they need from your stack, how to start and how to compare vendors.

TL;DR: what AI agents handle in an MSP

Work area

What the agent does

When it hands over to an engineer

Triage

Reads every incoming ticket, filters spam, sets category, priority and queue, links related tickets

Low confidence on classification or routing

Support resolution

Investigates and resolves routine end-user issues such as password resets, account changes, access and known software problems

Result cannot be verified, or the fix needs context the agent does not have

System administration

Executes administrative changes in Microsoft 365 and endpoint tooling within an approved action plan

Change falls outside the approved plan or touches high-impact systems

Proactive maintenance

Follows up on recurring alerts and maintenance tasks without waiting for a ticket

Alert pattern is new or the remediation is not known

What an AI agent for MSPs is

An AI agent for MSPs is software that completes service delivery work on its own: it decides what a request needs, acts through the MSP's existing tools, verifies the outcome and documents it, escalating to an engineer when it cannot finish with confidence. The defining trait is that it acts and closes the loop.

The word "agent" is overloaded in IT, so it helps to separate it from the tools it is often confused with.

Tool

What it does

Who decides and acts

Chatbot

Answers questions from a knowledge base

Nobody acts; the user still needs help

Copilot

Drafts replies, summaries and suggestions for an engineer

Engineer decides and acts

Script or workflow

Runs a fixed sequence when triggered

Engineer decides; the script acts

RMM agent

Small program on each endpoint that reports data and runs commands

Whatever controls the RMM

AI agent

Reads the request, chooses the action, executes it, verifies and records

Agent decides and acts within set limits

An AI agent usually works through the RMM agents already installed on your endpoints, and takes over the engineer time spent deciding which command to run and running it.

What AI agents handle today

Triage

Triage is where most MSPs start, because every ticket passes through it and the agent changes nothing in a client environment. The agent reads the ticket and any attachments or screenshots, filters spam and auto-generated noise, identifies the client and contact, sets category, priority and queue, and links the ticket to similar past cases. Consistent triage also improves everything downstream: SLA clocks start correctly, reporting becomes reliable, and resolution agents receive well-classified work.

Support resolution

Resolution agents work the routine end-user queue: password resets and account unlocks, MFA issues, mailbox and group access, license changes, user onboarding and offboarding, and common endpoint problems. A good agent follows an investigation order that fits the incident type, so it checks the obvious causes first and does not widen a single-device issue into a network-wide hunt. It replies to the end user in plain language and only reports a fix as done once the outcome is confirmed.

System administration

Administrative work such as tenant configuration, policy changes and scheduled changes is higher impact, so agents here operate within an explicitly approved action plan and with credentials limited to that task. It is usually the last area an MSP hands over.

Proactive maintenance

Much of an MSP's repetitive work never becomes a ticket: recurring alerts, patch follow-up, disk space, stopped services, health checks. The same agent approach applies, with the agent acting on the alert or schedule directly.

At Aegentics, triage (Jamie) and support resolution (Sasha) run in production at MSPs, and system administration (Cameron) is in controlled rollout.

Pre-built agents or build your own

There are three ways to get agents into an MSP, and they differ mainly in who carries the build and maintenance work.

Route

What you get

Who maintains it

Fits when

Feature in your PSA or RMM

An agent for a narrow function inside that vendor's product

The vendor, within that product only

You run a single-vendor stack and need one function

Agent builder platform

Tooling to create one agent per action (create user, reset password, clear disk)

Your team designs, tests and repairs each agent

You have engineers to spare for building automation

Managed role agents

Ready agents that cover a whole role, such as triage or support, across your stack

The agent vendor; you steer where your procedures differ

You want working capacity without an engineering project

The builder route looks flexible, but each agent is one more thing to keep working as clients, tools and APIs change. Aegentics takes the managed route: its agents behave like the human role they cover, and MSPs adjust their behavior through configuration instead of building logic per task.

What agents need from your stack

An agent is only as useful as what it can read and what it is allowed to do. Before a trial, check four things.

Ticketing. The agent needs read and write access to the PSA so it can classify tickets, add notes, update status and reply to end users. Aegentics works with Autotask, ConnectWise PSA and HaloPSA.

Endpoint access. To act on devices, the agent works through your RMM. Aegentics supports Datto RMM, NinjaOne, Kaseya VSA, ConnectWise RMM, N-able N-central and N-able N-sight.

Identity and Microsoft 365. Most routine tickets involve accounts, licenses, mailboxes or MFA. Access usually runs through GDAP relationships with delegated, scoped permissions.

Knowledge. Client documentation, procedures and past resolutions tell the agent how each client is set up. Aegentics reads from IT Glue, Hudu and ITBoost, and keeps approved facts per client so it does not have to rediscover them on every ticket.

Documentation does not have to be perfect to start. Agents work with environments as they are, and better documentation raises the share of tickets they can close over time.

How to get started

  1. Dry-run on your live queue. Connect the agent to your PSA in observation mode. It processes real tickets without acting and posts an internal note on each one: what it would do, why, which action, how confident it is and which past tickets it drew on. Your engineers see its judgment on their own work before anything changes.
  2. Switch on triage. Once the dry-run notes match what your team would have done, let the agent classify and route. It touches no client system, and the cleaner ticket data helps everything after it.
  3. Add resolution category by category. Start with high-volume, low-risk work such as password resets and access requests. Everything else keeps going to engineers, with the agent's investigation attached.
  4. Expand on measured results. Add categories when production numbers support it.

Run the trial on live tickets rather than a sandbox or a filtered queue. A sandbox shows how the agent handles test data; the live queue shows what share of your actual work it can take over, which is the number that matters for the decision.

How AI agents are priced

Vendors use five pricing models, and they behave very differently as ticket volume grows.

Model

How it works

What happens as you scale

Fixed per agent

A monthly fee per agent or per service desk

Cost stays flat while volume grows; easiest to budget

Per managed client or seat

A fee per client, user or endpoint

Cost tracks your client base, not agent usage

Per outcome

A fee per resolved ticket or completed task

Cost rises exactly when the agent performs well

Token consumption

You pay for the language-model usage behind each action

Hard to predict; long investigations and retries cost more

Bundled in a platform tier

Included in a higher PSA or RMM plan

Looks free, but ties the feature to that tier and vendor

Whatever the model, ask for a worked example at your own ticket volume and a hard monthly cap where usage is metered. Aegentics uses fixed pricing per agent: Jamie (triage) is EUR 499 per month after a free trial, and Sasha (support resolution) is EUR 1,900 per month, with triage as a prerequisite for resolution.

Checklist: questions to ask any vendor

  1. Does it act, or only suggest? Ask which ticket types it closes without an engineer touching them.
  2. Does it cover a role or single tasks? If every action needs its own agent, ask who builds and maintains them.
  3. Does it work with your current PSA, RMM and documentation platform? Native integrations beat a new connector layer you have to keep alive.
  4. Can you trial it on live tickets without risk? An observation mode on your real queue is the most reliable test.
  5. What happens when it cannot confirm a result? The answer should be escalation to an engineer, never an unverified message to the end user.
  6. How does it learn your clients? Ask how it uses your documentation and past tickets, and how it avoids repeating its own mistakes.
  7. What does it cost at your volume? Get a worked example and, for metered pricing, a cap.
  8. Does it meet your governance requirements? Audit trails, access control and data handling deserve their own review; our AI governance checklist for MSPs covers this in detail.

Frequently asked questions

What is an AI agent for MSPs?

Software that completes service delivery work, such as triage, support resolution or system administration, by reading the request, acting through the MSP's PSA, RMM and Microsoft 365 tooling, verifying the result and documenting it. It escalates to an engineer when it cannot finish with confidence.

What is the difference between an AI agent and a copilot?

A copilot suggests and drafts while an engineer does the work. An AI agent decides and acts within limits the MSP sets, then reports what it did.

Which tickets can AI agents resolve without an engineer?

High-volume, well-understood work: password resets, account unlocks, MFA issues, mailbox and group access, license changes, onboarding and offboarding, and known endpoint problems. The share differs per MSP and is best measured with a dry-run on live tickets.

Is it safe to let an AI agent work in client environments?

It is when the agent works within an approved action plan, uses credentials scoped to the task, logs every step and escalates results it cannot verify. Destructive actions should be outside its reach.

Do I need perfect documentation before using AI agents?

No. Agents work with environments as they are and use whatever documentation exists. Better documentation increases the share of work they can complete.

Which PSA and RMM tools does Aegentics support?

Autotask, ConnectWise PSA and HaloPSA; Datto RMM, NinjaOne, Kaseya VSA, ConnectWise RMM, N-able N-central and N-able N-sight; IT Glue, Hudu and ITBoost; and Microsoft 365 through GDAP.

How long does it take to see results?

A dry-run shows the agent's decisions on your own tickets from the first day it is connected. How quickly you move from observation to live resolution depends on how fast your team gains confidence in those decisions.

Further reading

This guide covers the practical side of AI agents. For what they mean for the MSP business model, read our series on the industrialization of managed services: