The One Support Ticket Agentic AI Will Never Autonomously Close

SAP just froze most hiring, paused non-AI travel, and started squeezing supplier budgets — all to shovel cash into what it’s calling the “Autonomous Enterprise.” Meanwhile, a SAP-commissioned study found 83% of enterprises believe agentic AI will transform their organizations. Yet only 3% say they’re fully prepared for it.

Ten days ago, PwC announced agentic contact and service solutions built with OpenAI, complete with a dedicated Center of Excellence, promising to “reimagine customer engagement” through an agentic front office. On top of that, Gartner is forecasting 63% growth in AI platform spending this year.

The pitch behind all of it is the same: agents that resolve tickets end-to-end. No human in the loop. Ticket opens, agent reasons, agent acts, ticket closes. In other words, autonomy as the finish line.

But here’s what nobody in those press releases wants to say out loud: there’s an entire category of support ticket that agentic AI will never autonomously close. Not next quarter. Not with the next model. Never.

The ticket that breaks autonomy

It looks like this:

“The light on the unit is blinking red.”

Blinking red how? Steady blink? Double blink? Which light — the one on the front panel or the one near the ethernet port? Is the cable seated? Which cable? Is that the WAN port or the LAN port? The customer doesn’t know. In fact, that’s why they opened the ticket.

Or: “I installed the replacement part and it still doesn’t work.” Installed it how? Backwards, half the time. Wrong slot, the other half. The customer will swear on their life they did it correctly, and they will be wrong. Meanwhile, no amount of agentic reasoning over the CRM record will surface that fact.

This is the ticket where the problem exists in the physical world, so the entire enterprise software stack is blind to it. The knowledge base doesn’t contain the customer’s living room. The device logs don’t show the crimped cable behind the desk. And the ticket history doesn’t capture the hairline crack the customer never mentioned because they didn’t notice it.

The bottleneck isn’t reasoning. It’s perception.

The agentic AI thesis assumes the constraint on resolution is cognitive: if the agent could just reason better over the available data, it could close the ticket. So the industry keeps making the reasoning better. Bigger models, more tools, longer chains of autonomous action.

But for physical-world tickets, the available data is the problem. An agent reasoning flawlessly over a wrong or incomplete description of reality produces a flawless wrong answer. Garbage in, autonomous garbage out — except now no human is around to catch it.

Text logs and CRM fields are a lossy compression of what’s actually in front of the customer. After all, a frustrated person typing “it’s broken” into a chat window is not a sensor. The autonomous enterprise can automate everything downstream of accurate information. But it cannot automate the acquisition of information it structurally cannot access.

In short, autonomy stops where the digital record ends.

What actually happens to these tickets

Today, the agentic loop hits the physical wall and does one of three things, all bad:

First, it guesses — walking the customer through a scripted troubleshooting tree that doesn’t match the actual problem, burning thirty minutes of the customer’s patience before escalating anyway.

Second, it escalates to a human agent — who is just as blind, working from the same text, playing twenty questions with a customer who lacks the vocabulary to describe what they’re looking at. As a result, your escalation chain becomes a game of telephone, except the first player couldn’t see the board.

Or third, it dispatches — a truck roll to solve a problem that’s often a cable in the wrong port. So the “Autonomous Enterprise” generates $300 field visits because its agents can’t look at a blinking light. Gartner already expects companies to cancel 40% of agentic AI support projects, and this gap is a big reason why: buyers judge these projects on the tickets that matter, and the tickets that matter are disproportionately the physical ones.

Give the agent eyes

The fix is not more autonomy. Instead, it’s more input.

The customer is standing in front of the problem holding a device with a better camera than most inspection equipment from a decade ago. So the shortest path from “it’s blinking red” to resolution is to look at it — a live video feed from the customer’s smartphone, straight into the support session. No app download, no setup friction, just a link the customer taps.

Once the agent — human today, increasingly AI-assisted tomorrow — can see the device, the physical-world ticket collapses into an ordinary ticket. The agent can identify the blinking pattern. The mis-seated part is obvious in four seconds. The wrong port is visible. As a result, diagnosis time drops from a twenty-minute interrogation to a single glance. This is exactly the gap Viewabo exists to close: remote visual support that turns the customer’s camera into the missing sensor in the loop.

And notice what this does to the agentic roadmap. Multimodal models are getting genuinely good at understanding video. That means an AI agent that can watch a live camera feed has a real shot at resolving physical-world tickets. An AI agent reasoning over text logs never will. So if you’re serious about autonomous resolution, visual context isn’t a nice-to-have — it’s the prerequisite. You can’t automate what you can’t perceive.

The 3% who are actually prepared

Go back to that SAP stat: 83% believe agentic AI will transform their organization, yet 3% are fully prepared. I’d argue most of the 97% are unprepared in a specific way — they’re buying reasoning capacity while their hardest tickets are starved for perception.

So before you sign the agentic platform contract, pull your ticket data. Then count how many escalations, repeat contacts, and truck rolls trace back to “we couldn’t tell what was physically happening.” That’s the ceiling on your autonomy investment. Every dollar of agentic AI spent above that ceiling buys brilliant reasoning about a reality it cannot see.

The autonomous enterprise will close a lot of tickets. But the one where the problem is sitting on your customer’s desk, blinking red? That one gets closed by whoever looks at it first. So make sure your agents can.