Joel Hellermark is a Swedish technology entrepreneur and artificial intelligence (AI) pioneer who, at the age of 30, has emerged as one of Europe’s most prominent business software executives.
HR and finance software-as-a-service (SaaS) applications supplier Workday bought the company he founded, Sana, in 2025 for $1.1bn. The concept behind the acquisition was to combine Sana’s AI-based search technology and agents with Workday’s technology to deliver employee experiences that are proactive and personalised to them, as a “front door” to Workday.
Born in Malaysia, Hellermark grew up between Tokyo, Japan, and Lidingö, Sweden, and developed an early interest in AI. At 13, he taught himself to code by independently taking distance learning courses from Stanford University. By 16, he had founded his first AI-related company, a video recommendation service, and at the age of 19, he founded Sana Labs.
Upon Sana’s acquisition, he became the chief AI officer at Workday.
In his 2017 TED talk, “How artificial intelligence will radically transform education”, Hellermark argued for AI’s potential to revolutionise learning.
Workday’s annual flagship Rising events are due to take place in the final quarter of 2026. Computer Weekly interviewed Hellermark at the supplier’s London Elevate event earlier this year – what follows is an edited version of that conversation.
I started the interview by reflecting on my early days in business IT journalism, in the late 1990s. It was the heyday of “knowledge management”. People talked about the knowledge economy, often assimilating that to the “internet economy” as well. And you’d go to conferences where people would talk about organisations being like brains, and what that would mean for organisational design.
It struck me, listening to Hellermark, that what he’s been doing with Sana and what he is now doing at Workday might put technology behind some of that rhetoric about the knowledge economy from the late 1990s and early 2000s. So, I first asked him what he thought about that.
How would you compare and contrast the knowledge management ideas common in the late 1990s with what you are doing with Sana?
I think there have been a lot of ideas that have felt inevitable but took a long time to get done. And what was missing was language models, to a large extent.
If you think about the company brain idea, it’s obvious that that’s something you would want. But getting to that point of what I like to describe as a “Leonardo da Vinci in every employee’s pocket” – a system that has all of your company’s knowledge, all of the world’s knowledge, all of your tools – took LLMs [large language models].
And then, once we got language models, we found the way we had structured data was not quite in order. And that’s shifting now to what I would describe as company memory.
I think instead of doing real-time retrieval of knowledge, we’re going to build out a company memory, which is a verified policy and a knowledge graph that agents can work with.
So, I think the core ideas are identical to the early ideas, from the early era of software, but now I think they actually work. And LLMs are what have made the real difference.
Language models have this ability to reason over large corpuses of knowledge and synthesise that and make it more accessible.
For companies, the first era was: “Can I build an internal Wikipedia?” The second was: “Can I build an internal Google?” So a search engine [idea]. The third was: “Can I build an internal ChatGPT?”
Joel Hellermark, chief AI officer at Workday, speaking at London Elevate
And now it’s really: “Can I build effectively a team of [virtual] co-workers that can support you in solving any task?”
Leonardo da Vinci is well known as the last person to have all human knowledge in his head. I think this system [Sana] is the first system to have all of your company’s knowledge in your head.
Leonardo da Vinci was, as an individual, incredibly polymathic. I’ve heard you mention the idea a few times that we are witnessing the return of the polymath and the rise of the polymath, with AI. That could be very daunting for a lot of ordinary employees, don’t you think? I mean, we can’t all be Leonardo da Vinci, right?
I think it’s quite the opposite. I think it’s quite accessible. It was quite daunting historically, if you just think about the number of tools you would have to navigate and the amount of knowledge you would have to gather to solve a single problem. This is becoming a lot more accessible.
In the early era of computing, we spoke about “bicycles for the mind”, tools that could augment humans to solve problems that historically had felt intractable.
That’s basically what we’re bringing to life now. It’s this bicycle for the mind that can augment humans in solving problems.
I think it can perhaps be daunting that we’re expecting more of the team, but I think it’s going to feel incredibly empowering [because of] the way these tools are implemented.
Historically, you had to do something that touched code. That would have been very daunting. “Oh, I have to learn the syntax of this and so on. And now, all of a sudden, it feels very intuitive. You just describe what you’re trying to solve, and it generates the code for you.
The concept of AI as a user interface (UI), which I’ve heard from SAP and others as well as Workday – can you unpack that for me? I understand that the idea is UIs will go away. We won’t be logging into individual tools and copying and pasting, it will be more “ambient”.
I think enterprise software today provides you with dozens of systems that you have to learn. You have to copy-paste between them. And they’re not really mapped to how you work.
But if you think about how you work, you’re trying to solve a task. That task might touch data across multiple systems.
What is increasingly happening is you’re generating the interface on the fly for the task that you want to solve. And the software is becoming invisible. You shouldn’t have to think about what the underlying system is, what the interface of that system is, and so on.
You should just describe what you’re trying to solve, and then it will gather the right data and the right interface to solve that.
And some of this will be enduring, so you’ll have applications that you’ll come back to. Some of it will be throwaway interfaces that are generated just for that task. And some of it will also be just overseeing the work that your agents are doing for you.
Are we far away from that, do you think?
No, I think it’s just going to be a journey until that is the default for every single workflow. Now, 10% of work is accomplished in this way. But if you start solving the integrations, when you solve the generative UI bits and so on, it will start being the majority. And at some point, you will never log into an enterprise application again.
It seems to me that a lot of what you’re talking about is being “meta” to the actual work – so we’re not getting bogged down in individual tasks, routine, mundane tasks, but we stand above some of those workflows and think about them in a more humanistic and creative way. Is that right?
I think that you’re at a higher and higher abstraction level. And that’s been the trend line since the early days of computing.
You used to have rooms of people who basically did the calculations of a computer. And then you had people using calculators, and then those just became cells in a spreadsheet. And now we’re not even touching the spreadsheet anymore. The spreadsheet is too low-level for us. We just have the agents doing the spreadsheet, and we’re at another abstraction level. I think that trend will continue.
You will have an infinite set of agents running in the background solving these tasks and just looping you in when necessary.
I remember Aneel Bhusri, Workday co-founder and CEO, saying in a press briefing earlier this year that what keeps him up at night is the fact that so many HR and finance tasks will be automated away. So, what do you do with the people who did those tasks?
I think as long as humans are the bottleneck, humans will be more and more valuable. And so I think over the next years, we’ll see the opposite trend. We’ll see increased employment and increased wages as a function of humans becoming more and more valuable.
If you think about these domains where you’re documenting and making humans more productive, the value of those humans is also massively increasing. A year ago, a software engineer was worth maybe one-quarter of what he or she is today. Now we can have a software engineer who’s overseeing the work of 10 agents in parallel. But I can’t have those agents without the software engineer. So that software engineer is incredibly valuable.
I was also thinking about something you said in that same press conference, which was about the scope of tasks that will be more and more automated, going from tasks that go from hours to minutes to big, day-long tasks. And that the next horizon for you was to look at the automation of those.
I think the game of AI resets every 12 months now. And just because you won the previous game doesn’t mean you win the next, so you always have to be on your feet.
The previous game was the game of co-pilots. It was feeding the AI context about a specific domain, having it answer questions more easily, and so on.
Second was the early agentic era where you could have these systems do single-step tasks. Now it’s: “How do you build the infrastructure for these agents to solve tasks that take a company memory model?” You need much more explicit policies. You need the rails and workflows for this to run on. So the game is reset.
Changing gears, does your experience with Sana make you more hopeful for European startups more generally – because IT is, for the most part, an American phenomenon?
I think what we in Europe have been historically very good at is the ability to take emerging technologies and build delightful user experiences on top of them. And this is an era where I think the interface for AI is yet to be defined, the way you collaborate with AI is yet to be defined, and so on.
I think the game of AI resets every 12 months now. And just because you won the previous game doesn’t mean you win the next, so you always have to be on your feet Joel Hellermark, Workday
I also think that the values we build companies on are going to be an advantage. Technology companies have a huge responsibility to be thoughtful in how AI is deployed. And that’s something we’ve always been really focused on.
It goes back to the early era of computing, with the story of Marvin Minsky and Douglas Engelbart running into each other in the corridors of MIT.
Marvin tells Douglas: “I’m going to make computers conscious. I’m going to automate intelligence.”
He was a big proponent of the idea of automated intelligence at that point.
To which Engelbart replied: “You’re going to do all of that for computers. What are you going to do for humans?”
He was always an advocate for augmenting human intelligence. He wrote that legendary paper, Augmenting human intellect, and building [in Steve Jobs’s later phrase] “bicycles for the mind”. And I think that’s been the principle we build on.
I think people will not just buy software from companies based on the software – they will also do so based on the values.
European software will be high-quality, delightful software built with the right values. Silicon Valley has had more of a “move very fast and break things” mentality. But there are companies like Anthropic that are very thoughtful about this. And frankly, this might be one of the reasons that Anthropic is currently dominating. It is a function of their values that enables them to attract the best talent.
I just think it’s part of the core DNA of our societies, and so it comes very naturally to European founders.
Current figures on subpostmaster accounting shortfalls put into perspective the Post Office’s “staggering mismanagement” of discrepancies experienced by those operating its branch network prior to…
French hyperscaler OVHcloud has made its Cold Archive cloud storage service generally available. It offers cheap long-term storage with cost per GB as little as €0.0013 per month. Cold…
Microsoft has issued four remote code vulnerabilities fixes in the September 2024 Patch Tuesday update, which it has marked as critical, meaning exploits are in…
Disgraced former Post Office CEO Paula Vennells sent out a company-wide email to counteract a Computer Weekly investigation that revealed the problems subpostmasters were experiencing…
