Dr. Vanja Josifovski is something of an anomaly, or as he puts it, “fortunate,” in that he has known what he wanted to do “since he was little.”

While those aspirations may not have been as direct as becoming the CTO of two of the biggest online brands of the last decade - Airbnb and Pinterest - that is certainly where he ended up.

The firms have become part of the fabric of modern life for many, and are part of the newer era of companies that arose alongside the advent of cloud computing.

When Josifovski was growing up, the concept of an online platform generating an income with no real tangible assets would have been hard to conceptualize.

Josifovski was raised in former Yugoslavia and his first interactions with “technology” came from his father having worked with Digital Equipment Corporation (DEC) in the 1980s. DEC is now part of Hewlett-Packard after a series of Matryoshka doll-style acquisitions.

Looking back at this time, Josifovski tells DCD that he can vividly remember the first DEC machine he saw: “It was PDP (programmed data processor) level, and it was still using cards.” Needless to say, the somewhat abstract concept of the cloud is diametrically opposed.

Originally, Josifovski wanted to be a professor. He taught for a year in Sweden. But, after doing a one-year sabbatical at IBM, ended up straying to “that route of industrial research that was very popular in the early 2000s.”

Starting at IBM Research, before moving to similar roles at Yahoo and Google, Josifovski was working in the place where “a lot of the current technology started, like transformers and foundation models, and so on.”

He says: “Suddenly, I woke up, and I realized ‘I don’t want to do research anymore’. You are publishing papers and talking to colleagues, but in some ways, science actually moves at a pretty slow pace. If you look at the technical papers, year after year, they are similar in structure, and the concepts and actually the math behind them do not move as fast as it looks from the outside.”

The frustration of academia and research is an oft-lived experience by those who dedicate their life to the pursuit. It is all-consuming and can, at times, lack the satisfaction of reaping tangible rewards like growing revenue or profit margins at a business.

It was this exit from research that was the axis upon which Josifovski did a hard pivot, landing as CTO of Pinterest.

Along with Jure Leskovec and Lance Reidel, Josifovski launched a software startup - Kosei - that was acquired by the social media platform after just nine months.

Pinterest itself was still a relatively new company, though it had grown rapidly. Founded in 2009 and launching with a closed beta in 2010, by 2012, it had 11.7 million unique US visitors. Josifovski’s business was acquired in 2015.

At the time of the acquisition, media coverage noted that there was a certain degree of uncertainty with some of the terms - a TechCrunch report quoted then head of engineering Michael Lopp as saying the Kosei team would be spending its first 90 days at Pinterest “figuring out where to apply themselves,” adding that it was unclear exactly what they would work on.

This was a sentiment repeated by Josifovski when speaking to DCD. “We negotiated for weeks, but we never found a good time to negotiate our roles, and we landed at Pinterest without set positions,” he explains with a smile.

“At the time we jumped in, it was a very small company, maybe fewer than 200 people working there. It was never my intent to be the CTO there. It’s a very different mindset from Google and the research side of large companies. When you go to small companies, everything is so imperfect. You have this assumption that everything has been talked through, but most of it has not. You have this assumption that people are more aligned than they are.

“But the reality is, especially with fast- growing popular companies, as Pinterest was at the time, it attracts a lot of very, very smart people, but then everybody comes in, and they try to put their stamp on the company. Everybody's pointing in a different direction, and they're all viable paths for success, but you need to figure out a path so they can all pull together.”

To say that those first 90 days of “figuring out” were something of a whirlwind would be an understatement, and certainly took Josifovski by surprise. “When I arrived, I thought I was going to do a lot of scaled machine learning - that's why they bought our company and from our backgrounds, but I quickly realised it wouldn’t make any sense, the data just wasn’t good. There was too much lower-hanging fruit to improve performance.”

Josifovski explains that, amongst this controlled chaos, he “went to Ben [Silbermann, co-founder of Pinterest] and asked, ‘What is the number one thing you care about today?’ And he said, ‘international growth’.”

He continues: “I left, and on the way home, I Googled ‘growth’, and I made a major proposal on the train. I presented it to him and the head of engineering, and three weeks later, I was head of growth engineering.”

Before long, that evolved to the CTO role at the platform, where Josifovski remained until Pinterest went through its IPO in 2019, at which point he decided he “wanted to do something different, and I was looking for a company where I would get out from the ad-powered model.”

Thus, the move to CTO of Airbnb came to fruition.

Airbnb and Pinterest, while having fundamentally different business models, do share one thing in common. Both were developed in the heat of the cloud computing revolution and have been cloud native since the get-go.

“They were both cloud native for practical purposes,” Josifovski says. “They were able to build and scale quickly because of the cloud offerings at the time. I would say that one of the last companies I knew of that built its own data centers in the Valley was probably LinkedIn, at the edge of the cloud computing revolution, in 2004 or so.”

Both Airbnb and Pinterest were based on Amazon Web Services at the time. “When I started, both were AWS,” says Josifovski. “They were probably in the top 10 customers and some of the early adopters of the technology. Now, they probably wouldn’t even break that top 10 because even large banks are heavily using the cloud.”

Part of his role as CTO at both was negotiating the cloud contracts Dr. Vanja Josifovski- attempting to get the best deals possible for the compute needed for the respective companies. That negotiation is a delicate dance.

“It’s an interesting trilemma, working out how to work with these companies,” says Josifovski. “You need to decide what you buy, what you build, and which cloud provider you buy from, and there are lots of pros and cons.”

But also, the CTO says, you “need some leverage.” He explains: “If you are single-cloud and a large buyer, you can get special contracts with discounts. These days, there are third parties that help with those things, but then it would be up to me to figure out the trade-off.”

A particular example of this came during Josifovski’s time at Pinterest.

“We would make a credible threat of moving out a percentage of our compute to another provider,” he says. “At some point, I would speak with AWS twice a year, and AWS would use those chances to present you with new products they have that might interest you. I would also start with a thirty-minute detailed and itemized plan of how we were going to migrate to a different cloud.

“It had to be credible enough to make sure we get the best terms. Of course, those terms are shrouded in NDAs, but there's always a little bird whispering somewhere. Those birds know a lot, and you can use that information to figure out how you can get the best pricing possible without going multi-cloud.”

This purist approach to cloud computing feels almost archaic at this point, with many companies today truly adopting hybrid or multicloud strategies. But in reality, this just goes to show how rapidly the market has evolved.

“It was a lot harder at that point to be multicloud,” he says, noting that today there are all sorts of platforms that can interact “automatically,” which offers a lot more opportunities. The major cloud providers today have also been quick to adopt solutions to enable this, with various interconnects between hyperscalers, and notably Oracle’s Database@ offering that sees Oracle hardware going into its competitor’s data centers.

Some of this is due to the industry in general being more evolved - cloud providers have found certain niches that they perhaps excel in, and thus have to accept that for some offerings, customers will want to adopt multiple solutions.

Another focus that has arisen over the years is a growing concern surrounding data sovereignty, and this was particularly pertinent for Josifovski while at Airbnb.

“When I was at Airbnb, the world started to require data to reside where it was used, and we went through a very difficult time trying to understand how to do this,” he says. “These companies that grew up in the cloud era, their software stack is like a living thing. It’s not like a Toyota Camry, where you can just turn it off and switch it back on again. It’s almost built to never be turned off, like an internal flame.

“Sometimes, turning it off can actually cause catastrophic failure, because it is evolving over time, and you can be sure it will start again. Because of this, it is almost impossible to take, say, Airbnb and just shift it somewhere else. There might be hundreds of services, and even if it’s the same cloud, you might have to start again from scratch.”

While a logistical nightmare, this also opens the door to doing things better and “from a clean slate.” As Josifovski puts it, it’s also an opportunity.

Josifovski left Airbnb in 2022, similarly shortly after its IPO, and made the move back to co-founder and CEO of a startup. These days, he has gone almost full circle, reuniting with, at this point, long term friend Jure Leskovec to launch Kumo. AI, a platform that uses breakthroughs in graph transformers and relational deep learning (RDL) to help users build predictive and embedding models.

Did Josifovski make this jump because he foresaw the AI mania currently engulfing the world? The short answer is, no.

“Google had some of the best people in the world, and I ate lunch with those people, and I didn’t know this was going to happen,” he says.

“No one sat at that table said this would happen, and certainly not that quickly. I was truly very surprised when prediction models started exhibiting some semblance of intelligence.”