AI for IT service providers and MSPs: what to automate first in 2026

The "what stays human" column follows the NCSC's MSP and prompt injection guidance (sources 1 and 2).

A search assistant over runbooks, client documentation and closed tickets answers technician questions in plain English and links each answer to its source. The technician checks the client's live configuration before acting on it. Documentation platforms often hold passwords next to procedures, so the assistant indexes procedures and never credential fields. The guide to AI knowledge assistants explains how search over company documents works.

Monthly client reports

The build drafts each client's report from PSA and RMM data, with ticket volumes, SLA results, top issues, patch status and open risks. It copies every number from the source systems. The account manager adds the commentary and any recommendation before the report goes out.

An AI tool inside an MSP can reach many clients' systems at once, so it gets the least access each task needs, and nothing it reads can grant it more. Three NCSC documents and one government bill set the standard clients will expect.

• Choosing a managed service provider (MSP), 24 November 2025. The NCSC tells small businesses to apply "least privilege (giving your MSP only the permissions needed to do the job)", to check that MSPs protect their own access to client systems with two-step verification, and to look for Cyber Essentials Plus. It also asks whether the MSP keeps security logs and whether clients can see them.
• Prompt injection is not SQL injection, 8 December 2025. The NCSC says language models "simply do not enforce a security boundary between instructions and data inside a prompt" and that "there's a good chance prompt injection will never be properly mitigated". It advises deterministic safeguards, keeping privileged tools away from any model that reads outside email, and logging inputs, outputs and tool use.
• Cyber Essentials. The scheme covers five controls: firewalls, secure configuration, security update management, user access control and malware protection. IASME says that for assessment accounts created after 26 April 2026, MFA is mandatory on every cloud service where it is available, and cloud services "cannot be excluded from scope". An AI provider account fits IASME's definition of a cloud service, which is "an on-demand, scalable service, hosted on shared infrastructure, and accessible via the internet".
• Cyber Security and Resilience Bill. The government's factsheet, updated 3 November 2026, says the bill will bring medium and large MSPs into the NIS Regulations 2018. They would need appropriate and proportionate risk measures and would report significant incidents to the Information Commission. Small and micro firms are exempt, and the managed services in scope include "IT remote support or helpdesks".

Controls for an MSP's AI build

• Separate API credentials for each client and each tool, scoped to the task, with no admin rights.
• The agent that reads inbound email cannot run scripts or change accounts.
• Action allow-lists and hard limits in ordinary code, with no deletes.
• MFA on every cloud account the build uses.
• A log of every prompt, output and tool call.
• A technician approves every change to a client system.

Client data and sub-processors

An MSP usually processes its clients' personal data as a processor. The ICO says "A processor may not engage a sub-processor's services without the controller's prior specific or general written authorisation", and processors remain liable for their sub-processors. An AI provider that processes ticket text containing client personal data acts as a sub-processor, so check each client's data processing agreement before connecting one.

Yarigai is a done-for-you AI agency in London. It builds AI agents, support agents that answer from a client's own documents, inbox triage, internal tools that replace spreadsheets and finance dashboards. For an MSP, it builds them inside the MSP's own PSA, email and documentation accounts. The MSP owns the code, the data and the accounts from day one.

A project starts with a free 30-minute scope call on Google Meet, and the MSP keeps the notes whether it hires Yarigai or not. A blueprint of about a week, credited to the build, sets out a written plan, a clickable prototype, the monthly running cost, a fixed price and the payback maths. The build takes 2 to 6 weeks with a demo of working software every week. A single workflow, such as ticket triage, can go live in about two weeks.

A person approves anything that sends money, emails a customer or deletes data. AI and hosting bills are paid by the MSP at cost. Yarigai's guarantee is one sentence: "If the build isn't profitable for you, we won't build it."

Yarigai has no published MSP case study. Its support agents and operations work are described on the services pages.