AI's value will come when we know it can do work for us, according to UiPath's Ed Challis

Last week, UiPath co-founder and co-CEO Daniel Dines (who will shortly step down to become Chief Innovation Officer, while retaining his controlling stake) gave an upbeat interview to FT-backed Sifted. In it, he announced that the firm aims to tap the benefits of generative AI, and perhaps make acquisitions in that space, arguing: 

We see [the use case] for generative AI around personal space and personal productivity. We think we can bring it into automated end-to-end processes....To make it a reality, it’s really hard work. Generative AI is fragile, hallucinatory, and automation needs reliability and predictability.

A refreshing, pragmatic view – especially in a market where other companies have talked up AI in the most hype-jumping terms, sometimes to ramp up their own share prices. Even so, the core message from Dines was simple: personal productivity is the future for RPA, and for UiPath.

The company then contacted diginomica and said it was keen to talk about interactive AI, in essence, AI that learns and you can have a more active, two-way relationship with. In our conversation, UiPath’s Senior Director of Engineering, Ed Challis, told me:

Ultimately, the mission statement of AI is technology which can automate tasks which require thinking, that require cognitive labor. But generative AI tools, they’re very much stuck in a personal productivity zone. That means they can help you write an email, they can help you write some code, but fundamentally, their value is capped at just helping you do stuff.

So does that mean personal productivity is old hat, and not the future the CEO said it was 24 hours earlier? This seems somewhat confusing. Challis continued:

The true value of this technology will be unleashed when we know it can do work for us. And that's a huge difference. I mean, imagine hiring someone who refused to look at anything, or to login into any system. Or refused to type…and all they would ever do is just talk to you.

Hey, I used to work with that person! But Challis seemed to be suggesting that generative AI is like that recalcitrant employee, rather than your helpful ‘know-it-all’.  Then he reiterated his point:

AI tools are very much stuck in that personal-productivity, advisory pattern of engagement. But their true value will be unlocked when we can plug them into our systems and get them to actually do work for us, without having to go through us as intermediaries.

For context, our conversation was sparked by recent comments by DeepMind founder Mustafa Suleyman – in the MIT Technology Review and elsewhere – that generative AI is just a “passing phase”, and the future belongs to interactivity. 

For his part, Challis was just getting started. He continued:

For AI to be truly valuable, it needs to have the same rights as an employee. It needs to have system access; it needs to be able to take actions…But, with people, you need to have a mechanism of trust, right, around your [human] employees and what they do. But, how do you gain trust in those people? 

First, they have to have the right background, and then they have to provide evidence. And then you need to go through an onboarding process. And they typically have a supervisor who reviews their work. And then, over time, you build out an evidence base that they will actually do the right thing and be consistent.

So, was Challis saying that AI will be more trustworthy than messy, complicated, expensive people? Not quite. He was suggesting that companies will need to onboard AI equally carefully. He explained:

Today, people typically copy and paste data in to GPT. But, you know, a really intelligent system lives in a loop, which is perception, action, then learning from those – from the outcomes of those perception/action loops.

The public perception of this technology is very much as a kind of chat tool. But the real value will come from being able to embed it in our processes and our systems. But in order to do that, we have to have real trust and verification that it's doing exactly what we want. That it's safe, and that it is compliant.

My take

On the earnings call following its recent Q2 results, analysts spent a lot of time drilling down on UiPath's AI strategy, both in terms of the tech, but also the monetization aspect.

The results showed revenues of $287 million, up 19% year on year – beating analyst estimates. The key focus of its financials was Annually Recurring Revenue (ARR), or Annualized Renewal Run-rate: $1.3 billion, up 25%  year-on-year.  

On the downside, UiPath recorded yet another quarterly loss, this time of over $60 million, two years of loss-making quarters on from its $1.3 billion IPO.

Over the course of the year to date, UiPath’s shares have ridden more peaks and valleys than the Carpathian Mountains, but overall, they are up over 39% - great news for investors, which is why it is a trending stock.

On the other hand, there is the matter of a flurry of ambulance-chasing announcements from US law firms last week to remind the market that an investor class action is building against the company, alleging securities fraud. How this plays out remains to be seen - such class actions are hardly unusual in the litigious US climate and their merits or otherwise need to be played out in the appropriate fora.  

At the time of writing, UiPath’s shares – up 39% in the year to date, remember – remain lower than those quoted in the action’s timeframe, at just over $17. This values the company at $9.7 billion – a significant sum, but only one-quarter of its market cap on the first day of trading (nearly $36 billion).

Aside from the financial stuff, UiPath certainly looks poised to learn from a perception/action loop of its own making.  More to come on the firm’s strategic direction - and messaging -  next week when it hosts its Forward VI event in Las Vegas, from which Alex Lee will be reporting.