Who builds custom AI agents for MSPs?

Custom AI agents for MSPs are built by three kinds of providers: PSA vendors that ship triage features inside their own product, MSP-focused AI products that plug into the common PSAs, and custom development firms such as Rubatt that build an agent around one MSP's own workflow. Which one fits depends on whether your problem is standard ticket triage or a process that only your service desk runs.

This page explains the three options, when custom is the wrong choice, what drives cost, and what to ask any vendor before you sign.

What are the three ways an MSP can get an AI agent?

| Option | What you get | Fits when | Watch for | | --- | --- | --- | --- | | Built into your PSA | Triage features switched on inside the tool you already use | Your need is categorizing, prioritizing and linking tickets | Features can be tied to a plan tier and a vendor roadmap | | MSP-focused AI product | A packaged agent that connects to your PSA and chat tools | You want a faster start and standard workflows | Per-seat or per-ticket pricing, limited control of prompts and logic | | Custom development firm | An agent built around your process, in your accounts | Your workflow spans several systems or has rules no product covers | You need a clear scope, an owner on your side and a way to measure it |

The built-in route is real and worth checking first. Datto describes Smart Ticket Triage in Autotask as automating "the categorization, association and routing of tickets based on historical data", and says it is available with Cooper Copilot in the Ultimate edition of Autotask PSA (Datto, Smart Ticket Triage). The same article describes automatic assignment to the most qualified technician as a later phase. That is a good example of the trade: little setup, but the scope and timing belong to the vendor.

What does "custom AI agent" mean for an MSP?

An AI agent is software that reads incoming work, decides what the next step should be, and drafts or performs that step using your tools. Ticket triage means reading a new ticket and setting its category, priority and owner. A PSA (professional services automation tool) is the system an MSP uses for tickets, time and billing.

A custom agent is one where the reading, the rules and the actions are written for your service desk. Examples of workflows that suit a first build:

1. Triage that uses your own client tiers, contract terms and escalation rules. 2. Drafting the first reply and the internal note, with the technician approving before it is sent. 3. Pulling context from documentation, the RMM and past tickets into the ticket before a person opens it. 4. Flagging tickets that look billable outside the contract, for a person to confirm. 5. Weekly client reports assembled from ticket data, reviewed before they go out.

When is a custom agent the wrong tool?

Often. If your need is plain categorization and your PSA plan already includes it, turn that on first. If a packaged product covers your workflow and its price works for your ticket volume, buy it. If the process is not written down, or two technicians would handle the same ticket differently, fix the process before anyone automates it.

Custom makes sense when the work crosses systems that no single product connects, when your rules are specific to your contracts, or when you want to own the logic instead of renting it.

What drives the cost of a custom agent?

We do not publish a flat price, because the same five things move the number on every project:

• Integrations. Each system the agent reads from or writes to adds work. A PSA with a documented API is easier than a tool with none.
• Write actions. Reading and drafting is cheaper to make safe than changing records or sending messages to clients.
• Evaluation. Testing the agent against a set of your real past tickets takes time and is the part that tells you whether it works.
• Data handling. Where client data is stored, who can see it and what a model provider receives.
• Running costs. Model usage and hosting continue after the build, and they scale with ticket volume.

Starting with one workflow keeps all five small. That is how we work: one workflow, shipped and measured, then the next.

What should you ask a vendor before you sign?

• Who owns the code, the prompts and the test set when the contract ends?
• Which actions can the agent take without a person approving them?
• Is every action logged, and can we see the log?
• How was it tested on our tickets before it touched live ones?
• Where does client data go, and which model provider sees it?
• What happens to our tickets when the agent is unsure?
• Can we switch an integration off without a rebuild?

Our own answers: clients own the code, prompts and evals. The agent reads and drafts, a person approves anything that cannot be undone, and everything is logged. More detail is on our pages about approval gates, integrations and security and data.

Do the rules differ in the US, UK and Canada?

Yes, and tickets usually contain personal data about your clients' staff, so this matters. This is general information and not legal advice. Check your own position with a qualified adviser.

UK. The ICO's guidance says "overall accountability for data protection compliance lies with the controller", and that a party is likely to be a processor if it only acts on a client's instructions. It also says a DPIA (data protection impact assessment) must be carried out before the processing of personal data begins. The ICO notes that this guidance is under review because of the Data (Use and Access) Act (ICO, accountability and governance implications of AI).

Canada. The Office of the Privacy Commissioner has published nine principles for generative AI, including consent, appropriate purposes, accountability and safeguards, for both developers and organizations using these tools. It tells organizations to know that accountability for decisions rests with the organization (OPC, principles for generative AI).

US. There is no single federal privacy law covering this. What applies depends on the state and on your clients' sectors, and your client contracts often set stricter terms than the law does. Read those first.

In all three, the practical step is the same: know which client data the agent will see, write it down, and tell clients where your contract requires it.

How does Rubatt work with MSPs and IT service firms?

Rubatt Inc builds custom AI agents and workflow automation for small and mid-size businesses in the US, UK and Canada, including MSPs and IT services firms. Our team is in Pakistan and works hours that overlap US, UK and Canadian business days. One example of our work in this area is a Microsoft Teams app for a Swiss cybersecurity firm.

If you have one service desk workflow in mind, see custom agent development or book a call. We will tell you plainly if your PSA's built-in feature or a packaged product is the better choice.