What "agentic AI" actually means for MSPs (and where the human stays in the loop)

"Agentic AI" has become one of those phrases that means everything and nothing. For some vendors it means a chatbot with a slightly longer memory. For others it means "we call an LLM API somewhere in the product." Neither is what we mean when we say Nexus is agentic by default.

Here is the specific claim: every module in Nexus — helpdesk, security, compliance, onboarding — has an AI agent that can draft a response, a diagnosis, a resolution, or a next step, on its own initiative, without a human asking it to first. That is the "agentic" part. The part that matters just as much is what comes next: a human always approves before anything actually sends, executes, or closes.

Why draft-then-approve, not autonomous

It would be technically straightforward to let an AI agent close tickets, run remediation scripts, or place outbound calls without a checkpoint. We deliberately did not build it that way, for a reason that has nothing to do with the current state of the models and everything to do with what IT operations actually is: a domain where a wrong action has real cost — a script that reboots the wrong device, a call placed to the wrong contact, a ticket closed before the fix actually landed.

So the design constraint is simple and non-negotiable: an agent can look at everything, reason about everything, and draft anything — a proposed fix, a suggested reply, a prioritized list, a call script. It cannot take an action with a side effect outside its own draft without a human clicking approve.

Every agent drafts and recommends. A human always approves before anything sends, executes, or closes.

What this looks like module by module

• Helpdesk: the agent reads a new ticket, checks device and site history, and drafts a diagnosis and a reply — a tech reviews and sends.
• Security: the agent prioritizes vulnerability findings by real exploitability, not raw score, and proposes a remediation plan — a human schedules and executes it.
• Onboarding: the agent walks a runbook for a new client, new site, or new hire and hands back a checklist — a human works the checklist.
• Calling: the agent can place a status-update or renewal-reminder call from a script it drafted — the script and the decision to place the call are both reviewed first.

None of these capabilities are live in production today — Nexus is in active development, and we would rather tell you that plainly than dress up a roadmap item as a shipped feature. What is fixed, and won't change as capabilities ship one at a time, is the supervision model above. See the full build sequence on the roadmap page.