AI Apps Revolution: The Rise of Smart Software and Investment Challenges

Rise of the AI apps

It's a truism in tech that every new computing platform opens the door to an entirely new generation of software companies. The client-server era that took off in the 1990s brought Oracle and SAP, while cloud computing gave birth to Salesforce and a host of "software as a service" companies. Large language models are shaping up to be the next platform to launch a thousand entrepreneurial dreams. With generative AI available on tap from companies such as OpenAI and Anthropic, there has been a blizzard of "smart apps" designed to make work easier. The speed at which some of these are winning users, and their surging valuations, is setting new records in the software world. Most notable has been the rise of coding assistants such as Cursor. Its owner is reported to be close to completing an investment round valuing it at $10bn -- only three months after it raised money at $2.5bn. Coding aids and other AI-powered tools for technically savvy users have led the way, but many other start-ups have been picking away at just about every aspect of white-collar work. These range from tools used to create or edit all forms of content and digital media to ones that can handle deep research. Fuelling this is a fear on the part of many workers that if they don't learn how to use the tools they will miss out on skills that will soon be an expected part of the job, says Tomasz Tunguz, a software investor at Theory Ventures. Some apps are registering surprisingly quick results. Mercor, which uses an AI-powered agent to carry out interviews to screen candidates for jobs, said in January its annualised recurring revenue hit $50mn less than two years after it was founded. For comparison, it took Salesforce four years to hit $50mn in annual revenue. Revenue at others appears to be exploding even more quickly. Loveable.dev, a Swedish company that tries to help non-technical users build things like websites, said its ARR hit $17mn last month, only three months after launch. A similar company, Bolt.new, said it went from zero to $20mn in two months. As companies like these achieve rapid lift-off, they face the same issues as generations of new software applications before them -- as well as a few new ones. One challenge is to turn an AI-powered tool designed for one task into a core part of a customers' software. That means automating more aspects of the processes they have targeted until their agents are capable of digesting an entire workflow. In this, they are up against software giants such as Microsoft, Salesforce and Adobe, which have their own AI agents and already have strong ties to many businesses. In the early days of the cloud, start-ups had a built-in advantage against incumbents, which had technology and business models tied to a different delivery method. But software's AI era is really more an extension of the cloud than an entirely new computing platform, points out Byron Deeter, a veteran software investor at Bessemer Venture Partners. That reduces the disruptive potential. Another difference has been the red-hot growth; this has made the most successful newcomers look more like consumer apps than traditional enterprise software, says Deeter. It isn't clear yet whether they will retain consumer-like characteristics as they mature, for instance leading to higher churn rates than are typically seen in the business software world. The financial profile also looks very different. The AI-app companies face a significant cost of goods, in the form of fees paid to LLM companies each time their services are used. Many are choosing to swallow those costs for now in the hope that LLM usage fees will continue to plunge. Cursor, for instance, lets customers make 500 calls a month for a subscription fee of $20, a price that is likely to leave it with little gross margin if paying full usage fees. Futurist and venture investor Peter Diamandis compares the huge investments being made in LLMs to the over-investment in new communications networks that occurred in the early days of the internet in the late 1990s. Now, as then, he says, the companies building the new infrastructure will be forced to slash prices and struggle to make a return, opening the way for the makers of applications to profit. The soaring tech valuations at the end of the 1990s ended in the dotcom bust. This time around, some of the app makers are at least generating serious revenue -- though that's no guarantee against another bubble forming as AI expectations jump.

"Read through the noise": Investors try to keep pace with AI

A large amount of venture capital has been invested in companies developing large language models (LLMs), which power AI tools like ChatGPT. As it gets harder for AI tools to stand out, the focus is shifting towards digital "agents" designed for specific tasks. "There are tons of companies out there," said Lauren Kolodny from Acrew Capital. "The challenge is trying to read through the noise."Two years into the generative artificial intelligence (GenAI) era, it's harder than ever for venture capitalists to spot winners in a rapidly changing competitive landscape. "This is a once in a generation opportunity" for investors, Sapphire Ventures president Jai Das said, during a panel at the HumanX AI conference in Las Vegas. A huge chunk of venture capitalist money has funded companies crafting large language models (LLMs), the software engines running AI tools like ChatGPT. But as it becomes increasingly harder for AI tools to stand out, attention is pivoting to digital "agents" capable of specialised tasks. "There are tons of companies out there," said Lauren Kolodny from Acrew Capital. "The challenge is trying to read through the noise." Early investors need to figure out which startups have sustainable competitive advantages, said Kolodny. Fen Zhao, director of research at Alpha Edison, believes it is "about how good a business you're making, not too much the underlying technology". Brian Goffman, from consulting firm McKinsey & Company, draws a parallel with the "SaaS" (Software as a Service) businesses that boomed with the shift to cloud computing. Once the remote cloud infrastructure came alive, corporations were flooded with lower-cost options. Standing out then -- as AI startups need to do now -- relied heavily on "identifying a business problem" and being the top solution, said Goffman. Gone are the early GenAI days, where having the most innovative model was the key. "There will be a lot of tears in some of these companies that end up not having the business model, but having some great technology team", warned Goffman. Entrepreneurs must establish where they fit in the big picture regarding the industry and the market, said Tomasz Tunguz from Theory Ventures. 'Steamrolling' Some investors question whether fending off rivals is even achievable in an ever-changing GenAI landscape. Josh Constine, from SignalFire, said it was not just about "first mover advantage". What also mattered is "proprietary data, and having the experts in-house to be able to train on top of that data". He pointed AI-powered platform EvenUp -- in which Signalfire has invested -- that caters to personal injury lawyers. It draws on a repository of prior settlements to provide guidance for new cases. "Companies that are building their own proprietary data pools are going to be the ones that are the most successful", Constine said. But James Currier, founder of venture firm NFX, challenges that view. "With 95 percent of the cases I can synthesize your data, I can copy your data", he added. Currier argues that companies should embed their product into the workflow of clients, creating a network that users rely on. "It's not going to be long lasting unless it has the sort of human-centred designer's focus on making the app actually work into people's workflows", Alpha Edison's Fen Zhao said. Established GenAI players are powering business tools like virtual sales associates, that follow up on leads, talk to customers and set up meetings. Constine said it reminded him of a decade ago, when startups relied on big tech platforms that were also potential rivals. "If you built an app on top of Facebook, that was too close to what Facebook's core mission was, you risked them suddenly launching a feature in the same space with all of their distribution advantage and steamrolling you," Constine said.