Sometimes, in looking at the AI startup landscape, you get a glimpse of how AI-driven business ideas are working out in the real world.

For me, there’s the sense that those watching the field closely have revised their outlooks over time, where in, say, 2021, many had a vague idea that “everything would be roboticized” and we’re now getting more of a granular perspective.

I’m thinking of a recent example where a trio of Harvard Business School students, also part of the Y Combinator community, talked about some of their applications to practical verticals like HVAC. This was at the Imagination in Action event in April, held at Boston (I’m responsible for helping to put this on) and we had some rousing conversations about everything attached to AI.

In this particular segment, Ben Horowitz interviewed Vidhan Bhaiya of Cortex, and Joe Schwartzman of Robbie. Cortex takes the aggressive position of acquiring various companies to inject AI into operations, where Robbie focuses specifically on aiding HVAC technicians in the field.

Initially, the group was talking about deployment speeds.

“What's happened to software has to happen to hardware,” Bhaiya said. “Hardware innovation cannot take 12 to 18 months to design every small thing, and we think that the missing piece in allowing that to happen is actually engineering judgment.”

So that explains why Cortex is acquiring firms.

Riding Along on a Furnace Call

Robbie, on the other hand, builds AI agents for home services technicians.

“We decided to just ride along with HVAC technicians and see what we could do with AI to help them with their day-to-day jobs,” Schwartzman said, going over the genesis of the company. “We pivoted around a bit and we're lucky enough to be in the most recent YC batch.”

Schwartzman then narrated a project that he said was not ultimately successful, to outfit HVAC technicians with smart glasses.

“This was a great HBS idea,” he said. “It sounded great on paper, and it didn't work in practice. When we put the glasses on technicians, it wasn't actually solving the problems that were most pressing to them. There was no clear link to revenue. There's honestly that disconnect between something that sounds nice, and a great business idea. So after spending more time, we obviously pivoted around and decided to go software-only first, which, to be honest, the underlying problem could have been solved with software.”

Later, Bhaiya talked about a specific way forward in aligning development to incentives.

“Incentives are just very hard to align,” he said, noting the importance of building trust, “and to make these tools valuable, you need to capture judgment and reasoning from these senior engineers.

That said, he delineated three steps to advancement:

“Step one is educating various people who are at very different parts of their AI journey,” he said. “I think step one to building trust is making sure that everybody comes up to the same level. So that's part of the transformation. Step number two is that we do believe that the people who are doing their tasks, the people who've been doing, say, a particular scoping task, or a quoting function and so on, they've been doing this for the last, say, 20-odd years. They are the best people in the world to tell us what the ideal state of this process is. We see AI, we see us as enablers to bring AI, to take them to that level. But we don't come in saying ‘this is what’s ideal’ - we ask them once we help take them there.”

“Step three is: can you share with us how you think about problems?” he explained.

Fear of Redundancy: Job Displacement Issues in HVAC and Elsewhere

In response to questions about job loss, the two sort of shrugged off some of the doom and gloom that you might hear around the water cooler.

“Engineers are actually not very motivated by money,” Bhaiya said. “Engineers actually want to do innovative work. They want to be pushing the boundary of innovation.”

He also distinguished between more generic processes that might take away jobs, and more surgical changes that focus on what AI is good at.

“We’re in pursuit of white spaces that otherwise would just not have been accessible with AI,” he said. “And that's something that's aligned with what everybody wants.”

For his part, Schwartzman pointed out that many HVAC technicians are “smug,” because they are less disruptible than others in other fields.

Toward the end of the talk, Horowitz asked the duo about innovating systems, inquiring: who is the audience?

Bhaiya suggested that the audience is those who are interested in rapid advancement.

“All the advances that have happened in software engineering, this is stuff that people are able to see, and we've just seen how software engineers are just able to do things that they couldn't do earlier,” he said.

At Robbie, Schwartzman said, the focus is on the technicians, the ones doing the work.

“Not many home service investors have actually been on the front lines, and that's kind of where the disconnect comes from,” he said.

Internal and External Strategy

I think that Bhaiya’s next point was very interesting.

How do you create influence as a third party?

Basically, Bhaiya started off saying that he would have preferred to just sell software, rather than taking over companies.

However, he said, the eventual strategy at Cortex, taking the reins at client firms, has created a lot of value. He explained.

“To capture data in context, we cannot do it as an outside vendor,” he said. “I think it requires incentive alignment, which means we have to be part of the inside.”

Cortex, he suggested, needed access to longitudinal data and the power to change business relationships in order to get the full value of the transformation. So they were left acquiring.

Then, Schwartzman went over how Robbie works.

“The data exists,” he began. “We surface valuable insights within that data. We make recommendations based on, for example, the expected value of an appointment. We translate that to the technician, give them a heads up, give them full context on everything they need to best serve that customer. And then after the appointment, we have a voice agent which collects all the information from the visit, to go back into the original CRM system.”

“We're understanding better how to increase revenue from every single visit,” he added.

Robbie’s approach, Schwartzman said, also solves problems for HVAC techs who have to do a decent amount of paperwork, albeit often on iPads, as they look around a project site for the first time.

“We found even though they're not our customer, they're not paying us, it is at least as good as what they're doing now,” he said, of technician input. “And actually, after they've started using it, it's better.”

As the talk concluded, Horowitz asked the other two about their outlooks on what propels the AI race forward, in general.

“This is the fastest adopted consumer technology in history, but people just forget about it as soon as they enter the workplace,” Bhaiya said.

Schwartzman suggested that, in the runup to today’s business strategy world, people were overcomplicating startups. He touted the influence of Y Combinator to focus founders.

“YC was great at just saying, ‘no, the only thing that matters is revenue, and the only reason why that matters is people are only going to pay you if they find what you're doing valuable,” he said.

Also, the two agreed in terms of simplification being central to new efforts. Essentially, in hashing this out at the end of the segment, there was the consensus that common-sense standards can drive startups, and adoption, and engineering.

I thought this was one of the better parts of the event, in a time when startups and even more established businesses dearly need orienting advice. Stay tuned for more.