February 21, 2024

Make way for one-person unicorns: AI startups say their software will replace thousands of jobs

“There's millions of people out there who have product ideas who just haven't had the time, money or energy to bring them to life.”

Tim Smith

6 min read

VCs have become addicted to investing in SaaS companies. With just two ingredients — a great idea and a great technical team — these startups could quickly and relatively inexpensively build  a big, impactful and valuable software business.

Now, some AI founders think they’re about to automate away one of those two magic ingredients — the people — potentially paving the way for unicorn-scale companies to be built on a fraction of the resources.

“The real problem businesses need solving is not having a software engineering team… We think that the user interface is going to shift away from having to hire to just interacting with an AI system,” says Anton Osika, founder of Lovable — a Stockholm-based AI platform that says it can let anyone build software applications “with just a conversation in plain English”.


Osika believes programs like his will be able to build “80% of all SaaS” software by the end of 2025, and before long, “you will see software unicorns where there is pretty much no human in the loop — it’s quite likely it will be just one person”.

But while some, like Osika, believe we could be moving towards a future when any individual can build the next billion-dollar software company single-handedly, others remain sceptical about whether AI is up to the challenge.

Some even think it could sink the entire VC model altogether.

Teaching AI to code like a human

If computer code is the DNA of a software business, GenAI models like ChatGPT have historically struggled to understand how the whole genome operates. 

Osika describes AI systems as being “blind alien artefacts” which, when given a programming task, can only see the exact job in front of them. What that means is that they might be good at writing small chunks of code, but struggle to build more complex products which might be interdependent on other systems.

Lovable’s product — which began life as an open source coding companion tool called GPT Engineer — is designed to make AI code more like a human does.

“What a human needs to do is to break it down into steps and, between each step of their plan, verify that the code they wrote is working as expected,” Osika explains. “It [Lovable] can’t do everything a human engineer can do but it can do much more than if you just ask it blindly: ‘Can you write this entire program?’”

For now, the founder says that Lovable is more suited to entrepreneurs who want to quickly build software prototypes — rather than a polished finished article — and can reduce the cost from around €3k to €300 for a functional early version of their product. Users can simply type what product they want into the interface with natural language, and the system will output working code, which can then be tweaked by an expert for finishing touches.

‘Total gamechanger’

One entrepreneur who is already seeing big potential in Lovable’s product is Ash Barbour, founder of Stockholm-based — a company that helps other businesses to bring the latest AI technology into their workflows.

He says he’s already helped a US real estate company to build a functional software tool for email triaging using Lovable’s technology, which non-technical people wouldn’t have been able to make in the past.

“It's really going to be a total game changer for non-technical founders,” Barbour says. “There's millions of people out there who have product ideas who just haven't had the time, money or energy to bring them to life.”


He predicts that we’re around five years away from software development being fully automated, but says that AI tools can already largely automate sales and customer service functions, with accounting being the next “low-hanging fruit” in the technology’s firing line.

“The more junior roles will be most likely to be automated first. A lot of companies are now just not hiring junior devs,” says Barbour. “You don't need someone to write your marketing copy when you can have GPT-4 do it.”

“That general trend is going to make it so hard for new grads and young people to get jobs, but could, in turn, encourage them to be entrepreneurs.”

Osika adds that, particularly in large, process-driven enterprises, AI tools like these will be able to do “what you needed thousands of people to do in the past”.

The bear case

While AI evangelists might be keen to talk up the potential for technology to allow anyone to run their own company without hiring legions of employees, you get a slightly more pessimistic view when speaking to entrepreneurs building businesses on their own.

Peter Nixey is the founder of UK-based time and task planning tool and says that, as a one-man-band company, he won’t invest in AI tools “unless he’s really convinced it’s useful”. He adds that, as AI founders promise that a bot can do the job of 10 junior sales reps, he’s yet to see meaningful results from tools that seek to automate a whole job (something he calls “outside-in” AI).

“There's this ‘outside-in’ [approach] where you create agents to just do a lot of stuff, and I don't see empirically interesting stuff coming out of that,” Nixey argues. He believes AI systems that address very specific pain points, like meeting summarisation or email management tools, are where value lies today. 

“Those more specific solutions: I would expect them to dominate for the next few years.”

When it comes to building software, Nixey thinks that AI models aren’t powerful enough yet to automate the process of product development.

“The AI is just not capable of doing it at the moment,” he says, echoing the sentiments of other founders who believe the “promise of these large models has been way ahead of the delivery”.

The end of VC?

While it remains in the realm of speculation, Nixey does see a possible future where next-generation AI models are able to build software products more proficiently.

If such a future is around the corner, it’s possible that the whole VC funding model could be upended, argues Barbour. He foresees software startups will find it harder to sell to bigger players, if software is so much easier to build in-house.

“How do you compete as a startup against the incumbents if they can build any software they want at the click of a button? The moat for SaaS is disappearing at a very drastic pace,” he says. 

Barbour says we’ll naturally begin to see more entrepreneurs developing vertical-specific tools to try and build more defensibility into their product, through deeper understanding of more niche markets, but that this will naturally lead to smaller exits.

“They [VCs] are predicting that a lot more founders will be having these sort of $50-100m exits, rather than $1-10bn exits,” he says. “And that's because it's going to be a lot more of this vertical play.

“The VC model doesn't work in its current form if founders are going for a $50-100m exit.”

Barbour believes that political interventions like universal basic income will be necessary if large companies can automatically build products that would have taken hundreds-strong teams years to develop, as huge swathes of existing jobs are wiped from the market.

Meanwhile, it will be people with strong “distribution and brand” who rise to the top of the food chain in an era of automatic software development: “My hot take is that influencers will be the entrepreneurs of the future; they have distribution.”

Better get posting on TikTok, folks!

Tim Smith

Tim Smith is news editor at Sifted. He covers deeptech and AI, and produces Startup Europe — The Sifted Podcast . Follow him on X and LinkedIn