Last year, DeepMind cofounder Mustafa Suleyman proposed a new type of Turing Test (a way of testing whether an AI has equalled or surpassed human cognitive abilities): can an intelligent system make $1m on its own?
It’s an idea which also speaks to one of the most alluring visions of an AI-powered future, where humans no longer have to work because machines are able to create economic value all on their own.
A growing number of startups are now trying to make these autonomous money-making companies happen.
Enter the AI agent
Building an autonomous company relies heavily on a type of AI known as “agents” — the basic concept being that you take all of the abilities of a powerful system like GPT-4, but you give it agency. So, rather than having to prompt an AI every time you want it to give you an output, an AI agent allows you to feed it one big task, and then go off and actually do the steps required to get there.
By way of an example, meet Alice. Alice is an outbound sales development representative (SDR) with one goal in life, generating leads to pass on to her colleague, who’ll take the first meeting with the potential customer. After a little training as to what your business does and who its typical customer is, she’ll go out and scour the internet for potential individuals or organisations to work with, and then reach out to them with personalised messages based on their profile and role.
Alice of course isn’t a real person, but one of four co-called “digital workers” developed by London-based startup 11x. Founder and CEO Hasan Sukkar tells Sifted that Alice is already able to outperform human benchmarks when it comes to successfully converting a lead into a meeting.
“It's not actually difficult to be better than the average worker,” he says. “The early feedback from customers is that this has tremendous ROI. One of our customers, they're using Alice at scale, they had 10 people running their SDR function in a way that Alice is doing single-handedly.”
Not the finished article
11x says it's currently got 50 paying customers and an even longer waitlist of companies who are interested in employing its digital workers. But despite the early appetite, Sukkar says that Alice and her AI teammates do still have some shortcomings.
One of these is due to the issue of “hallucination” — where GenAI models create output that’s inaccurate or otherwise not in line with the intended outcome.
“For example, I was reviewing something last week with one of our customers, where Alice had reached out to someone and she was like, ‘Hey, I noticed you've been an accountant for 30 years, that's very impressive.’ A human would probably not say this,” explains Sukkar.
While the 11x team is constantly working to refine its AI agents’ behaviour, Sukkar says it’s issues like this that make it hard for an AI to outperform the top 10% of performers, as there’s “an element of creativity and connectedness that's difficult to emulate”.
He adds that cost is still a limiting factor for widespread adoption of AI agents, due to the resource-intensive nature of running powerful AI models.
“We have a voice-based product that, to run it end to end, will cost $12 an hour on our side, and sometimes this is more expensive than the minimum wage in certain countries,” says Sukkar.
Increasing the market for AI agents
11x is just one of a growing number of startups building AI agents to sell into the digital workforce, with others including Sydney-based Relevance AI, Silicon Valley-based Lindy AI and Miami-based AirOps.
But none of these companies, which are generally automating specific, isolated job functions, are able to promise what you could call a fully automated company just yet.
One company that’s trying to address this and create a kind of company-wide system for controlling various different types of agents is London-based Honu.
Founder Imad Riachi describes the technology as a system that “understands the problem space” of your business — i.e. the market you’re operating in, your competitors and other elements of the problem the company is trying to solve — and orchestrates AI agents accordingly.
“It creates a common language where intelligent agents or applications can collaborate to optimise and solve the bigger business problem,” he says.
Honu is getting ready to roll out its automated decision-making system to its first customers, and Riachi says it will be able to work with a wide range of AI agents on the market. He adds that, rather than just doing what it’s told and carrying out human-decided tasks, Honu will actually be able to suggest ways to make your business better.
“It might sound esoteric but it’s like the ecosystem will be conspiring to make your business work better,” says Riachi. “If you've got $1,000 to spend, it’ll tell you where your money is best spent.”
A new economy
Riachi predicts that by next year we’ll see the first fully autonomous profitable business, saying it’s likely to be something like a small ecommerce company that gets there first.
“It’ll be businesses that have a simple enough business model, that are highly digitised and have an interesting ecosystem of services around them that you can leverage to be able to make that business successful,” he says, adding that the role of CEO will soon shift to being someone who knows how to direct these AI systems.
Sukkar says that agents like those created by his company could very well be used in cost-saving measures — such as mass redundancies — and that as a society “we need to be smart with how we think and operate in this new reality”.
Both founders repeat the mantra that all businesses are likely to need fewer people due to this next wave of automation, but that the number of successful companies will greatly increase.
“When you build a decision infrastructure that uses AI to optimise the business, you're making the bar to entrepreneurship much lower, so business creation goes through the roof and (business) default rates plummet,” he says.
In many ways, the rise of AI agents in the workplace simply continues a trend that’s been going on for decades: software and AI automating away certain functions of our jobs.
But if autonomous companies do arrive in the next couple of years, society also might have to start asking itself what kind of businesses we want to create — and whether an economy that’s powered by an ever-expanding number of ecommerce sites is one we want to build.
