ConnectWise's RMM agent diagnoses the device behind a ticket, then waits for a person. The wait is the design
ConnectWise launched an agentic RMM that diagnoses endpoints read-only and stages fixes from a vetted catalog, plus intake agents for voice, Teams and email. All four are generally available.
Listen to this article · 5 min
AI-generated narration of the full article.
At two in the morning, a ticket arrives: a user’s laptop cannot reach the file share. A technician who picks it up at eight will spend the first twenty minutes doing what every technician does. Check the device, check the network, check the logs, try the obvious things, and only then start thinking.
ConnectWise’s launch on October 29 automates those first twenty minutes and deliberately stops there. Its new Agentic RMM puts an agent inside the remote monitoring and management platform that managed service providers use to run their clients’ endpoints. When a ticket comes in, the RMM Agent investigates the device behind it, tests the likely cause and stages a fix from a vetted catalog. The diagnostics are read-only. Any change to the device needs explicit approval from a technician, given inside the ticket.
ConnectWise also made three intake agents generally available, for voice, Microsoft Teams and email, which collect missing details from the end user before a ticket reaches a person. All four agents are described as generally available at launch.
Read-only first is not timidity
It would have been easy to market this as autonomous remediation. ConnectWise chose a narrower shape: the agent may look at anything, test hypotheses without changing state, and propose a change, but only from a catalog of scripts the provider has already vetted. That maps neatly onto the ladder we proposed for earning agent autonomy: read, then recommend, then act with approval. The RMM Agent sits on the third rung on purpose.
For an MSP, the reason is commercial as much as technical. A provider manages endpoints for many clients under contract. A wrong change made automatically on a client’s machine is a breach of trust with a customer, not just an incident. Keeping the technician as the one who presses the button keeps accountability where the contract puts it.
The vetted catalog matters as much as the approval. It turns “what may the agent do” into a list someone maintains. If a remediation is not in the catalog, the agent cannot propose it, which bounds the damage of a confident wrong diagnosis.
What partners report
The figures in the release come from ConnectWise partners and are anecdotes, not controlled measurements. RealBytes says its agents evaluated 99 percent of 7,074 tickets in 30 days and returned 178 hours to the team, and that more than 500 live tickets went through the RMM Agent, with a useful diagnosis in 91 percent of them, each in about a minute. Fusion Computing reports 86 percent fewer escalations. RealBytes puts breakeven on its AI spend at about 16 minutes saved per technician per day.
“Useful diagnosis” is the phrase to examine. It is not the same as a correct one. A team adopting this should define, before the pilot, what counts: the diagnosis named the root cause, the staged fix was the one the technician applied, or the technician merely found it helpful. Each gives a different number. The operations agents we reviewed from AWS and Azure raise the same point: a confident explanation is not evidence until someone checks it against what fixed the problem.
What to ask before turning it on
Who curates the catalog, and how is it versioned? Scripts drift. A remediation that was safe for one OS build may not be for the next.
What does the agent read? Read-only diagnostics can still touch sensitive data on a client’s device: file names, event logs, user activity. Clients may need to know.
How is the approval recorded? The approval inside the ticket should show what the technician saw: the diagnosis, the proposed script, the device. Otherwise “approved” becomes a reflex.
What happens to the hours? If the first twenty minutes disappear, the provider has to decide what the capacity is for: more clients per technician, faster response tiers, or deeper work. Without that decision, the saving stays on a slide.
The launch is a useful example of agentic IT done with restraint. The agent does the investigation people find tedious; the decision stays with the person who answers for it.