Agentic AI is revolutionising how startups are built. Just about every business process in every industry is now up for grabs as a new generation of founders frantically works out the best ways of deploying AI agents in the real world and building new businesses.
As the Sifted leaderboards of fastest-growing European companies show, startups are growing faster than before at lower cost, largely thanks to the smart deployment of AI. How is this going to change the tech industry’s investment calculus?
One person with strong views on the subject is Henrik Werdelin, the Danish entrepreneur and investor who has spent the past 15 years co-running the startup studio Prehype. With his business partner Nicholas Thorne, Werdelin has now launched a new venture called Audos, promoting a very different investment model by backing thousands of “everyday entrepreneurs”. To that end, Audos has raised $11.5m from investors, including True Ventures, Niklas Zennstrom and Leila Zegna.
Rather than trying to hunt down rare and elusive unicorns like a traditional VC, Wederlin aims to let loose herds of so-called donkeycorns (he may need a better marketing agency). The definition of a donkeycorn is that they should be run by a solo founder with an intense passion in a particular niche, whether it is expertise in Californian divorce law or post-partum fitness regimes. “They grind like mules, but they party like unicorns,” he says.
Werdelin believes that there is enormous pent-up demand among frustrated corporate employees to strike out on their own. Agentic AI is now giving them powerful tools to do so. According to a Harris poll from last year, 59% of employed Americans would like to start their own business but lack the money or knowledge to do so. “The TAM is not always big, but for that person with a very specific skill, they can now codify it into a business,” he says.
Wederlin’s eventual aim is to help launch 100,000 businesses a year with up to $25k of funding apiece. Each startup will have no more than 2 employees and will shoot for $1m of annual revenue. This mass democratisation of entrepreneurship could, he argues, create a new version of the Mittlestand, the SMEs that helped drive Germany’s postwar economic miracle. “I would like to live in a world where there’s a million people building $1m turnover businesses,” he says.
Rather than taking equity in the business, Audos will charge a 15% annual fee. That’s partly because Werdelin does not believe that many of these businesses will exit – and also because it will help Audos generate cashflow and back new investment opportunities.
As well as financing, Audos gives entrepreneurs access to the company’s business-building toolkit that was developed at Prehype and enables them to develop effective AI agents. This should help startups find rapid customer-market fit and expand fast, Werdelin says.
Audos certainly has an intriguing financial model and it will be interesting to see how its investments fare. For many founders, the perpetual 15% fee will seem steep. The most ambitious of them may also prefer to stick with the traditional VC funding model, which promises greater scaling possibilities and exit outcomes. The speed at which agentic AI is evolving may also give more autonomy to solo entrepreneurs, reducing the attractions of Audos’s business-building methodology.
As Emma Burrows, co-founder of Portia AI, told the Sifted podcast, it is “dizzying” how fast agentic AI is evolving. The tech sector is now expanding from a two-body ecosystem – mixing machines and humans – into a three-body ecosystem with the addition of agents. It is hard to understand the art of the possible and predict how this new tech universe will evolve. “It’s both been surprising how fast it’s moved, and also surprising how far we’ve still got to go,” she says.
The tech investment landscape of the future is unlikely to be dominated by just unicorns and donkeycorns. Who knows what new mythical creatures agentic AI will help create?
Read the orginal article: https://sifted.eu/articles/ai-agents-vc-donkeycorns-john-thornhill/
