Why Everyone Is Building the Same HVAC Chatbot

Picture a technician on a hot roof, one glove pressed against a rusting condenser, squinting at a metal nameplate that lists a model number nobody has thought about since 2009. He needs the wiring diagram. He needs the capacitor spec. For thirty years the answer was a phone call to a guy named Dave back at the shop. In 2026, the answer is a chatbot. And there are at least a dozen companies fighting to be the chatbot he opens.

That is the strange thing about this moment. The product everybody is building is nearly identical. Scan the nameplate, pull the manual, draft the job notes, suggest the fix. Nobody is claiming a secret. The pitch decks rhyme. What separates the players is not the feature - it is who already has the technician's boss on a monthly invoice.

So this is not a story about air conditioning. It is a story about moats. HVAC is just the smallest, cleanest room in which to watch a much bigger question play out: when a specialized AI product races an incumbent platform that already owns the customer, who wins?

The two campsDepth in one corner, distribution in the other

On one side are the AI-native startups, and they are betting everything on depth. NOSO Labs, out of Y Combinator's Summer 2025 batch, has spent its early life doing something deeply unglamorous: assembling a reported 20 terabytes of equipment manuals so a technician never hits a machine it hasn't read about. Its founders, Winston Chi and Alex Xi, are treating the manual library itself as the product. Prentis, still listed as "coming soon" with an August 2026 target, promises hands-free voice diagnostics across 146 makes - the kind of thing you want when both hands are inside an air handler and your phone is on the ground.

Then there is Avoca, which took the wedge in a different direction - answering the phones. In April 2026 it announced more than $125 million raised at a $1 billion valuation to run an AI front office for home-services companies. A unicorn built on picking up the call when a homeowner's furnace dies at 9pm. Around the edges sit OxMaint, an AI maintenance system aimed more at industrial facilities than the guy in the van, and Gooey.AI, a general-purpose AI platform that shows up on market maps of this race even though its HVAC connection is thin. Market maps inflate. Worth remembering.

What unites them is a bet that the technician's knowledge problem is real and unsolved. HVAC expertise has always lived in three places: the manual nobody can find, the head of the veteran who is about to retire, and the phone number of the one guy at the shop who has seen everything. That is a terrible way to run an industry, and it is a perfect target for a language model. The startups looked at 425,000 technicians carrying decades of scattered know-how and saw a dataset waiting for a wrapper. They were right that it was a dataset. The open question is whether the wrapper is a company.

On the other side are the platforms, and they are not really betting on AI quality at all. They are betting on the login. ServiceTitan unveiled Atlas, its AI "sidekick," at its Pantheon conference in September 2025. Atlas scans a nameplate and pulls manuals, wiring diagrams, and parts across more than 200 brands through a Bluon integration. It drafts job summaries. It helps dispatchers pick the right technician. If that list sounds familiar, it should - it is roughly the startup roadmap, shipped from inside software thousands of HVAC firms already pay for every month.

And ServiceTitan is not alone. Jobber launched its Copilot in October 2024. Housecall Pro rolled out an entire "AI Team" of role-based agents in the fall of 2025, including a 24/7 receptionist. FieldEdge is inching in with dispatch suggestions. And the enterprise giants have arrived with their own layers - Salesforce Field Service with Agentforce, and Microsoft Dynamics 365 Field Service with Copilot. Everyone with a field-service product has bolted an assistant to it.

The mechanismWhy this isn't a coin flip

Here is the part worth staking a claim on. Most "who will win" pieces end in a shrug - it depends, time will tell, may the best team ship. That is not useful, and it is not honest. This race has a named variable, and you can check it yourself in eighteen months.

The incumbent's advantage is distribution. ServiceTitan does not have to build a better assistant than NOSO Labs. It has to build a good-enough one and drop it into a product the customer already opens forty times a day. There is no new login, no new invoice, no procurement meeting. The AI is simply there, included, the week it ships. When you already own the relationship, "good enough and free" beats "excellent and separate" almost every time.

The startup's only counter is depth, and depth has to be the kind an incumbent cannot cheaply copy. More manuals. More brands. More obscure 2009 model numbers. Tighter trust with the technician, who can smell a tool that has never seen his equipment. And speed - the ability to ship a fix on Tuesday that a company serving thousands of customers needs a quarter to schedule. A big platform is heavy. Heaviness is the startup's one gift.

But there is a catch inside the catch, and it is the reason this race is genuinely hard to call. Depth is copyable if it is just data. ServiceTitan's Bluon integration already reaches past 200 brands, and an incumbent can license a manual catalog as easily as a startup can scrape one. So the durable version of depth is not the size of the library - it is the trust and the workflow wrapped around it. A technician does not switch tools because one has 20 terabytes and the other has 18. He switches because the tool answers the question he actually asked, in the two minutes he has before the customer walks back into the room. That last mile - the part that is stubborn, specific, and unglamorous - is where a focused startup can still out-build a platform that is trying to serve plumbers and electricians and roofers in the same release.

Why you should careThe technician is a stand-in for you

If you have never touched an air conditioner, this still concerns you, because HVAC is just the concrete version of an abstract fight happening everywhere. Legal AI startups are racing the platform that owns the law firm's document system. Healthcare AI point solutions are racing the electronic health record that every clinician already logs into. Insurance, logistics, accounting - pick a vertical, and somewhere a thin AI layer on an underserved dataset is discovering that the incumbent with the login just shipped the same feature for free.

HVAC is the cleanest test because the stakes are low and the picture is legible. A guy fixing an AC unit. A nameplate. A manual. There is no regulatory fog, no ten-year sales cycle to muddy the result. When the dust settles here, the answer will generalize - which is exactly why a piece like this becomes the thing people search for when they type "vertical AI moat" into a box a year from now.

It is tempting to treat the incumbents' AI push as a defensive shrug - bolt on an assistant, tick the box, keep the churn down. That underrates them. Distribution is not just a shortcut to users; it is a data flywheel the startups do not have. Every job ServiceTitan's customers close feeds the system that makes the next dispatch smarter. The platform sees the whole loop - the call, the schedule, the invoice, the follow-up - while a point solution sees one slice of it. If the incumbents ever get serious about quality, they are not starting from zero. They are starting from the richest operational dataset in the trade. The startups' window is real, but it is a window, not a door left permanently open.

So watch the depth. Watch whether NOSO's manual library grows faster than Atlas can absorb Bluon's catalog. Watch whether Prentis actually launches in August, and whether hands-free voice is a feature or a company. Watch whether Avoca's front office becomes so good at the phone that ServiceTitan would rather partner than rebuild. The incumbents have the customer. The startups have the head start and the heaviness of their rivals. That is a real race, not a foregone one, and the beautiful part is that we get to check the score.

One prediction, plainly stated so you can hold us to it: distribution wins the median case. The typical HVAC firm will use whatever assistant is already inside the software it pays for, and for most of them that is an incumbent. But depth wins the edges - the firms with unusual equipment, the technicians who care, the segments too weird for a platform to prioritize. The startups that survive will be the ones who found an edge worth owning and got so deep the incumbent decided copying wasn't worth it. Come back in a year. Bring the receipts.