Can an AI briefing prevent a second elevator service call?

NOVEMBER 3, 2026 · THE WEEKDAY BRIEF

Good morning. Before Torben Teichmann walks into a Düsseldorf office building, he can ask his phone which elevator needs attention and which tools to carry. For a service business, the most valuable answer might be the one that saves a second trip. How would his employer know what that answer was worth?

Today we follow a maintenance record from one technician to the next, then look at how Microsoft checks machine-written code and how NVIDIA tracks stalled computing jobs.

Torben Teichmann · Max Brunnert for Microsoft

In an October 22 account, Microsoft describes two tools TK Elevator gives its technicians. One retrieves manuals and service history before a visit. The other asks technicians to describe what they did and follows up when the record lacks detail. The next technician can start with a colleague’s account of what has already been tried.

Consider a sticking elevator door. Knowing which part a colleague replaced last month could save a trip back to the van. A plausible but inaccurate repair note could send the next person out with the wrong part. Both outcomes begin with a faster handover. Only one gets the elevator back into service sooner.

Here is an illustrative calculation for a team making 20 visits a day. Suppose dictation releases five minutes per visit. That gives the team 100 minutes. Allow two minutes per visit to check the record and the team gets 60 minutes back, before software and support costs. That hour may be scattered across several routes. A dispatcher still needs a nearby job and a technician with the right parts to turn it into another completed call.

Microsoft’s account documents the workflow. It gives no controlled estimate of savings. To price a pilot, compare similar routes before and after rollout and track repeat visits for the same fault. Include a comparison group with the same digital records and dispatch tools. Better recordkeeping alone could explain some of the improvement.

Read Microsoft’s field report

Microsoft’s migration review process · Microsoft Command Line, figure 2

Microsoft engineers Sriram Rajamani and Tim Bozarth describe a similar handover in software. A small team reviews rules drawn from a sample of code changes. Agents can then apply those rules across a much larger codebase. One expert decision can guide thousands of edits, which makes the quality of that decision especially valuable. A mistaken rule travels just as far. Count accepted changes per hour of expert review, including time spent reversing changes that passed the first check.

Read the engineers’ account

Illustrative server racks · Brett Sayles / Pexels

NVIDIA’s Minerva connects a computing job to the equipment serving it. If one storage drive slows a training run, an operator can trace the delay across the cluster. NVIDIA identifies IREN as a user. For an operator considering the system, the economic test is how many useful jobs finish on existing equipment after accounting for the cost of monitoring and repairs. A shorter list of alerts is helpful, but it won’t pay for a rack by itself.

Read NVIDIA’s Minerva announcement

What changed? A technician’s spoken report can inform the next visit.

What is measured? The workflow is documented. The 60 minutes above are hypothetical.

What still costs money? Checking notes, dispatching crews and returning to unresolved faults.

Who captures the gain? The service company can gain capacity. Tenants gain time when the elevator returns to service sooner.

What information would have saved your team its last repeat visit? Hit reply and tell us.

— Machines and Money team