The Office Got the AI. The Truck Did Not

The short version

Artificial intelligence (AI) adoption in the trades is real, and it went to the office. Three surveys break adoption down by business function, and all three lead with office and business work: quoting, invoicing, business writing, marketing, and administration. None of them includes a line item for a technician diagnosing equipment on a service call.

The gap is structural. Office tasks are text problems with clear inputs and outputs, already inside the software the shop runs. Field work starts from pressure readings and temperature measurements at the equipment, depends on the manufacturer and the installation, and happens where holding a phone is the third priority behind safety and the measurement. The homeowner who called about a unit that will not cool is still waiting on the same diagnosis whether the office runs on AI or not.

Where the adoption is

Three surveys published in 2026 break contractor AI use down by business function.

Jobber, a field service management platform, surveyed 1,050 home service business owners and reports these rates among those using AI: 54 percent for quoting, 52 percent for invoicing, 51 percent for business writing, 35 percent for customer communication, and 29 percent for scheduling and dispatch.

Houzz, a home renovation and design platform, surveyed 601 U.S. construction and design businesses and reports these rates among construction firms using AI: 64 percent for sales and marketing, 61 percent for planning and design, 59 percent for project and client management, 52 percent for administrative work, and 45 percent for business operations.

The Associated General Contractors of America (AGC), with Sage, surveyed 951 respondents from construction firms. Of those responding to the AI question, 45 percent report using AI for office and administrative functions, 23 percent for estimating, 20 percent for design or preconstruction, and 16 percent for recruiting and training.

The denominators differ: Jobber and Houzz report rates among owners and firms already using AI, while AGC and Sage report rates among all firms that answered their AI question. The columns cannot be compared directly, but the pattern is the same: office and business functions come first, and no line item covers a technician diagnosing equipment.

Function	Jobber	Houzz	AGC/Sage
Quoting and estimating	54%	—	23%
Invoicing	52%	—	—
Business writing	51%	—	—
Sales and marketing	—	64%	—
Customer communication	35%	—	—
Planning and design	—	61%	20%
Project and client management	—	59%	—
Administrative work	—	52%	45%
Business operations	—	45%	—
Scheduling and dispatch	29%	—	—
Recruiting and training	—	—	16%
Onsite activity and monitoring	—	—	13%

The one entry closest to field work is AGC and Sage’s “onsite construction activity, monitoring, and documentation,” chosen by 13 percent of firms. That option covers monitoring and documentation rather than diagnosis.

The clearest product story in this category is answering the phone. Jobber describes its Receptionist as a feature that answers calls and texts around the clock, captures job requests, and books visits during the call. Housecall Pro’s customer service representative AI (CSR AI) answers calls and chats, with text answering still in an alpha test. Both solved a problem every contractor recognizes. Neither diagnoses equipment.

Census Bureau researchers, working from the 2026 AI supplement to the Business Trends and Outlook Survey (BTOS), report that adopting firms use AI most commonly for sales and marketing at 52 percent, strategy and business development at 45 percent, and IT at 41 percent. The same researchers report that 57 percent of AI users integrate AI in three or fewer business functions. The adoption is narrow even inside the firms that have it.

The five-to-one ratio

Researchers at the Federal Reserve Bank of St. Louis, analyzing European firm-level data on AI adoption, report that “among European firms, the share using AI for any business purpose is, on average, five times larger than the share using AI specifically for production processes.” Among EU firms that deploy AI, 35 percent use it for marketing and sales, 31 percent for business process organization, 23 percent for accounting, and 21 percent for production processes. Production trails all three office functions.

The same pattern appeared in the Census Bureau’s own question. Before November 2025, the BTOS asked whether a business used AI “in producing goods or services.” After that date, the question asked whether a business used AI “in any of its business functions.” Reported adoption nearly doubled, from about 10 percent to 17 percent, and the Census Bureau started a new time series rather than continuing the old one.

What changed was not how many firms used AI. What changed was whether the question counted the use that was already there. The Census Bureau’s own cognitive testing on the original question found respondents who first said no and then turned out to be using AI provided as part of software workflows. The adoption was invisible to the people it belonged to, because it arrived inside software they already ran rather than as a product somebody purchased.

Office tasks were the first AI use case to reach broad adoption because they fit: text in, text out, inside software the shop already pays for. A technician’s work is a production process, and production trails the office functions.

Why the field is a different problem

Office work is a text problem. A quote has a template, a customer name, and a set of line items. An invoice has the same. A marketing email has a recipient and a topic. The inputs and outputs are words and numbers in a structured format, and the software the shop already uses holds all of them. AI that generates text from text fits directly into that workflow.

A technician at the equipment works from readings: suction pressure, discharge temperature, superheat, subcooling, amp draw, static pressure, return air temperature. What those readings indicate depends on the metering device. A fixed orifice system is commonly charged by superheat and a thermostatic expansion valve (TXV) system by subcooling, but the manufacturer’s procedure for that model and those conditions comes first: Lennox directs a weigh-in on some fixed orifice units. The same low superheat reading on the two is read against different expectations, and low superheat alone does not give you a cause. The answer depends on the manufacturer, the model, the refrigerant, and what the installation looks like.

The input is not a text prompt. It is a set of measurements that the technician took with a specific instrument, combined with observations about the physical system: the condition of the filter, the state of the coil, whether the blower wheel is loaded, whether the sensing bulb on the TXV is making full contact with the pipe. No software the shop already runs holds that information, because the information does not exist until the technician creates it at the unit.

The environment is the other half. Much of the work takes both hands on instruments, some of it happens where there is no cell signal, and some of it happens in heat that makes reading a phone hard. The phone is the tool of last resort rather than the primary workspace.

DEWALT surveyed 2,481 U.S. construction professionals in December 2025, including 1,730 skilled trade workers and 751 decision-makers across residential, commercial, and industrial construction. Ninety percent of respondents believe AI will be indispensable in their industry within five years. Eight percent use it in their day-to-day work.

The gap between 90 percent belief and 8 percent daily use is not skepticism. Eighty-seven percent say AI education should be built into trade school programs, and 59 percent want hands-on training tied to real tasks. DEWALT reports the primary barrier respondents cited is a lack of formal, job-relevant training.

Pew Research Center, analyzing occupation-level data in 2023, found that mechanical skills, including equipment maintenance, are more important in jobs with less exposure to AI. That finding matches what the data in this post also shows: the value a technician provides happens at the equipment, and the tasks at the equipment have not been the ones AI addressed first.

What exists at the equipment

Trane Technologies launched an AI-powered Tech Assistant in September 2025, integrated into the Trane Technician App and the American Standard Technician App. Paul Parish, Trane’s General Manager of Digital Residential HVAC, described the goal as “putting advanced AI capabilities directly into the hands of skilled technicians” to enable “faster problem resolution.” The tool searches Trane and American Standard product documentation and responds to technician queries in the field.

Documentation search is useful. A technician who needs a wiring diagram, a fault code definition, or a specification spends less time searching literature or waiting on a supervisor. The technician still writes the query, though Trane says the tool asks follow-up questions when one is unclear.

The question a diagnostic AI would answer is different. A technician with high subcooling and low superheat on a TXV system does not need to look up what those readings indicate in general. The technician needs to know which of the possible causes fits this unit given these conditions, and what to check next to separate them. That is not a document lookup. It is a reasoning problem: the tool starts from the technician’s own measurements and works through the diagnostic tree with them.

No publisher behind these surveys reports a rate for that kind of tool, because the category barely exists. What got adopted is office software with AI features built in. The field is where the technician shortage is sharpest, and it is the last place the technology arrived.

My take

We build a diagnostic AI for technicians at Prentis, so this post is about the gap our product is built to address, and you should read it that way.

The five-to-one ratio from the St. Louis Fed researchers shows why the office was served first and also why the field is harder to serve. A tool that writes a follow-up email can be built into an existing platform in months. A tool that works through a diagnostic on a TXV system with the technician’s actual readings needs to know the difference between a fixed orifice and a TXV, what each reading rules in and out, and when the most likely cause depends on something the technician has not measured yet. That takes years, and it is why technicians are still waiting.

Common mistakes

Read the function breakdown before the headline adoption rate, because the headline counts shops with AI somewhere in the building and the function breakdown shows which tasks it runs on.
Ask whether a tool works at the equipment or at the desk, because these surveys’ function breakdowns are led by desk work and the technician shortage is at the equipment.
Check what the AI feature inside your existing software actually does, because Census Bureau cognitive testing found respondents who first said no to using AI and turned out to be using it inside workflows they already ran.
Treat a documentation search tool as a lookup, because it starts from a query you write and Trane’s launch release describes documentation search rather than working through a set of readings with you.
Ask what a tool changes about the technician’s time at the unit, not about the office’s time on the phone.

Don’t

Do not read an adoption rate as meaning AI reached the field, because Jobber puts HVAC adoption at 81.5 percent while its survey-wide function breakdown leads with quoting, invoicing, and business writing.
Do not buy a field tool on the strength of an adoption rate for office software, because an industry rate covers what other shops told a survey and not what your technicians will open at the unit.
Do not assume a tool that answers calls also helps the technician on the call, because Jobber’s Receptionist and Housecall Pro’s CSR AI answer phones and book jobs and neither diagnoses equipment.
Do not dismiss AI for the field because no headline rate covers it yet, because DEWALT found 90 percent of construction professionals expect AI to be indispensable within five years and cited lack of job-relevant training as the primary barrier.
Do not confuse documentation search with diagnostic reasoning, because looking up a fault code definition and working through a set of readings to identify a cause are different problems.

Frequently asked questions

Is AI being used by HVAC technicians in the field?

AI use by HVAC technicians in the field is minimal as of mid-2026. The three contractor surveys that break AI adoption down by business function put office work first: quoting, invoicing, business writing, marketing, and administration. AGC with Sage reports the closest field-adjacent line item at 13 percent of firms for onsite activity, monitoring, and documentation, and that option covers monitoring and documentation rather than diagnosis. DEWALT, surveying 2,481 U.S. construction professionals in December 2025, found that 8 percent use AI in day-to-day work while 90 percent believe it will be indispensable within five years.

Why did AI go to the office before the field?

Office tasks are text problems with structured inputs and outputs inside software the shop already runs. A quote has a template, an invoice has line items, and a follow-up email has a customer name. Researchers at the Federal Reserve Bank of St. Louis report that among European firms, the share using AI for any business purpose is five times larger than the share using it for production processes. The U.S. data shows the same pattern: when the Census Bureau broadened its AI question from “producing goods or services” to “any business function,” reported adoption nearly doubled.

What AI tools exist for HVAC technicians in the field?

Trane Technologies launched an AI-powered Tech Assistant in September 2025 that searches Trane and American Standard product documentation to answer technician queries. Jobber and Housecall Pro each offer an AI feature that answers calls and books jobs around the clock, but those tools solve an office problem rather than a field one. Documentation search starts from a query the technician writes, and no tool cited here works through a set of readings and observations to help identify a cause.

What did the DEWALT AI study find?

DEWALT surveyed 2,481 U.S. respondents in December 2025, including 1,730 skilled trade workers and 751 construction industry decision-makers across residential, commercial, and industrial construction. The study found that 90 percent believe AI will be indispensable within five years, 8 percent currently use it in their day-to-day work, and 87 percent say AI education should be built into trade school programs. DEWALT reports the primary barrier respondents cited is a lack of formal, job-relevant training.

Glossary

AGC: the Associated General Contractors of America, a construction trade association that publishes an annual survey with Sage.
AI: artificial intelligence. In the surveys cited here, the term covers chatbots, language models, automated scheduling, and documentation search tools.
BTOS: the Business Trends and Outlook Survey, a Census Bureau survey of about 1.2 million businesses that carries the AI questions cited here.
CSR AI: Housecall Pro’s AI customer service representative, a feature that answers calls and books jobs.
DEWALT: a power tool brand under Stanley Black and Decker that published a study on AI adoption and training in the skilled trades in April 2026.
Fixed orifice: a metering device, such as a piston or capillary tube, that does not adjust refrigerant flow. Superheat is the common charging check on these systems, but the manufacturer’s procedure for the model and the outdoor conditions comes first.
Houzz: a home renovation and design platform that publishes an annual State of AI in Construction and Design report.
Jobber: a field service management platform for home service businesses that published a 2026 Home Service Trends Report.
Receptionist: Jobber’s AI feature that answers calls and texts and books visits.
St. Louis Fed: the Federal Reserve Bank of St. Louis, whose researchers analyzed how AI adoption measurement varies with question wording.
TXV: thermostatic expansion valve, a metering device that adjusts refrigerant flow based on suction line temperature. Subcooling is the common charging check on these systems in cooling mode, but the manufacturer’s procedure for the model, the mode, and the outdoor conditions comes first.

Drafted with AI assistance and reviewed by the author.

← All posts