After a year spent integrating ICT services provider Umbrellar into Pax8, Umbrellar's former CEO, Dave Howden, has reemerged leading AI startup SupaHuman.
After seven years at Umbrellar, Howden said he could see the wave of driving NZ businesses to the cloud was starting to flatline.
"We'd done our job," he said. "We'd done some of the largest migrations the country had seen to Microsoft Azure and we were pretty proud of what the company had done."
Having seen the cloud boom starting in 2015 and 2016 and riding that wave, the emergence of large language models and AI presented an opportunity to apply that experience to his own business.
A couple of years on the board of New Zealand AI engineering pioneer Ambit AI and the emergence of GPT technologies delivered the possibility to spin-off a new company to solve big engineering problems using large language models.
"The biggest learning I took from Umbrellar is you can't specialise enough in some of these emerging technologies," Howden said. "You have to be very laser focused.
"I couldn't see an organisation that existed today that was going to focus on engineering large language models to deal with specific New Zealand problems.
"That smelt like an opportunity for me to pull some of the best minds and the best talent the country has together to help organisations engineer those solutions."
Launched in January, SupaHuman, is focused on solving the AI challenges NZ businesses face and helping them to adopt AI safely and at velocity, Howden said.
The key to that was creating domain-specific large language models for local businesses.
Where tools like ChatGPT can be good at generic tasks, they don't know enough to perform more specific ones because they don't have access to proprietary data.
"Unless you put your intellectual property openly on the internet and give the AI the ability to focus its skill-set on your knowledge, then it's always going to be generic," Howden said.
Rather than focusing on the automation of specific manual tasks, which will mostly be achieved when existing software platforms are enabled with native AI tooling, SupaHuman aimed to support "next level gear-change thinking".
"Everyone is infusing AI with their products," Howden said. "The reality is, and this is a good thing, they are infusing their domain-specific knowledge into their products so they can continue to be competitive and take your money on a monthly basis,"
However, he said, if you could digitise all of the capabilities of an engineering team and put that into a single large language model, you would have the best, most productive employee that is always online, 24x7 to answer any form of engineering query.
That would allow real people to focus on the creative side rather than the programmic ones. For any organisation with substantial domain knowledge, human work would then centre around managing the AI and to be a custodian of the customer outcome.
Where SupaHuman comes in is to provide cloud agnostic advice and engineering to companies trying to solve gnarly questions.
"We build on top of Microsoft, AWS and OpenAI, all of the large language models," Howden said.
"As it stands right now we believe we are the only AI technology solutions provider in New Zealand. That is our narrow skill-set: we are not doing cloud migrations, we are not doing architecture, we are engineering AI solutions."
The business model is "agency-like", coming at customer challenges from a creative angle.
"What could you possibly imagine your business to be if you could digitise your intellectual property? What could that look like? And then we engineer that solution into life."
Howden also points to the need to manage intellectual property.
"If it's in people's heads at a time when labour markets are so turbulent with the movement of talent around different industries, you're going to see this labour arbitrage to AI quite quickly," Howden said.
"That can come across quite negatively, but the reality is it is those that embrace the digital versions of their intellectual property are going to win the race."
Add robotics into that and you have a whole different ball game.
"You're looking at knowledge arbitrage and labour arbitrage all at the same time," Howden said. "That's not our game because we are not in the robotics business, but we are in the digitisation of IP."
SupaHuman AI has a team of eight so far, predominantly engineers and all at PhD level.
"I do not profess to be an AI expert; I'm a business guy," Howden said.
"My job is to bring the talent together to engineer these outcomes and making sure we are connecting the business problem with the technology solution."
The goals were to ensure the new firm had the leading engineering capability in the country and to "move past the hype".
A local focus would enable AI to perform within the NZ context by producing ethical AI that aligned with NZ policy, law, regulations and the Treaty of Waitangi among other considerations.
"We think that's going to keep us busy for the foreseeable future," Howden said.
SupaHuman's go-to-market is not to deploy a technology solution but to define the opportunity for AI and then engineer that. It could, therefore, lead to brand new products or brand new business models.
As for the future of the channel, it's a question of how quickly it can pivot.
"What happened in cloud was a five to seven year cycle," Howden said. "We're going to see that same thing in under two years for the emergence of AI because of the speed at which it can create value."