Agents The diagnosis starts before the truck rolls
It is 6:50 a.m. in a light-commercial service office and the dispatcher is staring at a single alarm — not cooling — for a rooftop unit across town, with no readings, no history, and a tech about to drive out blind.
Source: ACHR News, "Q&A: How AI is Affecting Commercial HVAC Systems" — https://www.achrnews.com/articles/166758-q-and-a-how-ai-is-affecting-commercial-hvac-systems — the Q&A describes connected equipment reporting recorded readings and telemetry from when the alarm fired, compared against the unit’s own history, with AI analysing equipment behaviour to surface abnormal patterns and likely root causes before the tech arrives, shrinking the diagnostic portion of the visit and cutting no-problem-found trips. No vendor results are claimed in this post.
The problem
Most commercial calls still start with almost nothing: a broad alarm, a tenant saying it feels warm, and a unit installed years ago by someone else. The tech drives out and spends the first part of the visit working out what the equipment was doing when it failed — if it failed at all.
That blind start is expensive in the ordinary way. The wrong parts ride in the truck, and a share of visits end as no-problem-found because the fault cleared before anyone arrived. The customer paid for a diagnosis that started at the curb instead of at the alarm.
How the agent workflow works
- The alarm carries its evidence with it. When a connected unit faults, the workflow captures the readings and telemetry from that moment — pressures, temperatures, run state — instead of leaving the tech to reconstruct them later.
- Today is compared with this unit’s own history. The agent lines the alarm snapshot up against how that same unit normally runs, so a slow drift and a sudden break do not look like the same call.
- Abnormal patterns get named before dispatch. The tech’s brief lists what looks out of pattern and the most likely root causes to check first, in order, rather than a generic fault code.
- The truck gets loaded for that short list. Parts, gauges, and access needs follow from the brief, so the first trip has a real chance of being the only trip.
- A cleared fault still leaves a record. If the unit is running on arrival, the tech works from what it was doing at the alarm, not from a blank screen — the difference between a checked system and a shrugged no-problem-found.
- The visit closes the loop. What the tech actually found is written back against the prediction, so the next brief starts from a corrected history.
What you need to set it up
- Connected equipment on the accounts you aim this at first: units that actually report readings and telemetry, with enough sensing and variable operation for there to be something to analyse. Light-commercial accounts with connected gear fit; unconnected residential splits do not.
- Data organised before AI touches it. The source is direct: readiness starts with data readiness — performance records in a usable structure, tied to the right unit.
- Defined limits the agent cannot cross. Equipment protection — compressor limits, pressures, coil temperatures — stays with the equipment controls, inside set bounds toward the comfort target.
- Cost, honestly: the source publishes no price for this capability. Expect connected-controls or platform costs plus any sensor or connectivity upgrades a building needs, quoted per site — not a flat app fee.
Where it breaks
Unconnected equipment gives it nothing to start from. A single-stage compressor and a fixed-speed fan offer the controls one decision — on or off. Without richer sensing and variable operation, the brief is thin no matter how good the software is, and selling this workflow for that fleet is overselling it.
A likely cause is not a diagnosis. Pattern-matching can point at the probable fault and still be wrong, and a tech who treats the brief as the answer stops checking. The brief orders the first tests; hands, gauges, and judgment on the roof still settle it.
The learning curve is human. Crews used to setpoints now get ranked causes. If the office presents that as the system knowing better, techs will ignore it. It lands when the brief is a head start the tech is expected to overrule.
The bottom line
Start with connected commercial accounts you already serve, and measure one thing: how often the tech’s first checks match the brief. The win is not a smarter dashboard. It is a call that begins at the alarm, not again at the curb.