Your AI agent is only as good as the tickets it can't see
Every AI agent being sold into IT operations right now has the same blind spot. It can only act on what it can see, and the moment the work crosses from one company to another, it stops being able to see it.
That’s not my opinion. Gartner reckons that by 2030, half of all AI agent failures will come down to two things: weak governance and systems that can’t talk to each other. Interoperability is a top-two reason these things fall over. Not the model. Not the prompts. The plumbing underneath.
What this looks like on a normal Tuesday
An enterprise runs ServiceNow. You run ConnectWise, and you support them. Someone on their side asks their new agent a simple question: where’s this request up to? The agent checks the only system it can see, their ServiceNow, and answers straight away. Still open. No movement since yesterday.
Except your team picked it up first thing and closed it out before lunch, over in your ConnectWise. The work’s done. The agent didn’t lie. It just can’t see your platform, so it answered confidently off half the picture. And a confident wrong answer is worse than no answer, because nobody thinks to check it.
That’s the bit that gets me. The agent isn’t wrong about what it can see. It just can’t see half the conversation, and the half it’s missing is the half where the work actually happened.
This isn’t a corner case. Cross-company work is where most of the messy, high-stakes stuff lives, and it’s exactly where today’s agents go dark.
The vendors are building agents on one platform
The platform vendors know agents are what everyone wants, so they’re building them straight into their own stacks. ConnectWise into ConnectWise, Kaseya into Kaseya. All useful. But each agent sees one platform, the one it lives in. ConnectWise’s agent doesn’t see your client’s ServiceNow, because why would it. The intelligence is being built on top of the single-platform silo, which is the one place cross-company work was always going to fall through.
The protocols move messages, not matching records
You’ll hear the new protocols are going to fix this. MCP, A2A, agents learning to talk to each other. They’re real and they’re useful. But they standardise how agents talk, not whether two companies’ tickets actually match. A protocol carries a message. It doesn’t decide which platform is the source of truth, or map your ticket to their incident, or keep both honest with a log when they disagree. The first spec that even tries to join those two protocols up isn’t due until later this year, and it’s about messaging, not keeping records in sync.
Agents reason. Integration is infrastructure.
This is the bit people skip. Agents are good at reasoning. Integration is infrastructure. You don’t replace a pipe with a brain.
Before any agent can do something useful across a boundary, someone has to have already decided which system wins when they disagree, how the data’s governed as it moves between two companies, and how every change gets logged so it stands up later. That’s not thinking work. It’s a data layer. And right now it’s the part nobody’s funding, because it’s the boring bit.
The boring bit is what we build
The boring bit is what we build. Support Fusion keeps the ticket in your ConnectWise and the incident in their ServiceNow as the same thing, both directions, with an audit trail, so whatever’s reading the data, an agent, a dashboard, a person, is reading the truth. We sync the whole object, not three hundred field mappings that break the day someone adds a column. And we run it, so you’re not the one watching it.
We’re not an agent, and we’re not pretending to be. We’re the layer the agents need and don’t have yet.
The future everyone’s selling assumes the agent can see the ticket. Across the boundary, it can’t. Get that right first and the clever stuff on top finally has something true to stand on. Get it wrong and you’ve bought a very confident way to make your best tech look like they didn’t turn up.
See it move
Easiest way to judge this is to watch it. Give us half an hour and we’ll show you a ticket move across the boundary and close itself on both sides, on your platform and your client’s. Book a demo and we’ll set it up.
Nathan Tremlett is Founding Head of GTM at Support Fusion. www.suppfusion.com
Sources: Gartner, Top Predictions for Data and Analytics 2026 (March 2026), for the agent-failure prediction.
