As the hype surrounding AI agent startups continues to grow, VCs are preparing for a bust cycle that could wipe out many of the sector’s darlings.
Investors warn that the cycle is building because some companies developing agentic AI — large language model-based applications to automate tasks without a human in the loop — are raising money at inflated valuations and for products that might not live up to expectations.
Six weeks into 2025, European startups developing AI agents have picked up €481m, according to Sifted data — more than a quarter of the funds raised by such startups across 2024. Many expect investment to go up and to the right over the next year.
While investors scramble to back winners in the space, there’s an air of caution hanging over the sector. The startup world is no stranger to a hype bubble bursting on a new, and potentially transformative, technology, with recent examples in the swift rise and fall of a number of Web3 and metaverse startups.
“There is a significant risk of a boom-bust cycle with AI agent startups,” says Denny Gabriel, associate at Runa Capital. “This era is no different than the cloud, mobile and internet eras.”
The rise of AI agents
Agentic AI is promising to transform business with an array of use cases across sales and customer service to legal and healthcare. It's an alluring play for VCs because it ticks two boxes.
One, AI agents have the potential to reshape how businesses operate and could bring huge cost savings by automating work that would previously need to have been done by a human.
And two, the tech is typically low-cost to build — and expected to become cheaper over time as the price of compute, which is needed to train models, comes down.
“In the past two years, rapid advancements in large language models have pushed AI from simple, chat-based interactions to systems capable of advanced reasoning and decision making,” says Adam Said, investor at Octopus Ventures.
Since the beginning of 2024 Julia Flaig, principal at Join Capital, says there’s been a big uptick in pitches by startups saying they’re developing AI agents. “Almost all software companies are now positioning themselves as AI agents in some shape or form.”
Notable AI agent rounds in the past year include London-founded 11x, which raised $74m across two rounds in 2024 and relocated to the US, and Paris-based H, which raised $220m from investors including the US’s Accel. AI darlings like Mistral and defence tech Helsing are also building AI agents.
The frothy AI agent startup market
For most startups AI agent technology is still a work in progress.
“Building a one-size-fits-all agent to seamlessly plug into the hundreds of software and data sources enterprises rely on and orchestrate layers of actions is a big engineering challenge,” says Adam Shuaib, partner at Episode 1 Ventures.
Many startups pitching themselves as AI agents aren’t actually building truly autonomous agents capable of controlling whole workflows without human intervention, Shuaib adds.
“‘Agentic' has become a buzzword that is thrown around a lot despite not typically representing the actual capabilities of a product,” he says.
“One of the main things we’re noticing is the lack of definition over what these ‘agent’ systems really are, with a few too many ‘agent for agent-sake’ type pitches,” says Oliver Kicks, partner at Concept Ventures.
“A few AI agent plays are starting to ring alarm bells akin to the Web3 and Metaverse mania of recent times.”
Web3 startups raised more than $50bn globally in 2021 and 2022, according to Dealroom, more than 10 times the previous two years, before funding came tumbling down to $4.8bn in 2023 as crypto markets crashed. VCs similarly jumped into the metaverse sector around the time Facebook rebranded to Meta in 2021, before the downturn led to funding falling off a cliff in 2023.
Big opportunity
There’s also the challenge of scaling in a new market when there’s no playbook to follow.
“During these cycles, revenue models and value capture remain uncertain, making it difficult to assess long-term durability for a lot of these businesses,” says Gabriel. “In venture capital, this often results in many startups securing funding but failing to reach meaningful scale before running out of capital.”
“We remain particularly cautious of startups that cannot clearly articulate how they will transition from early experimentation to generating real, sustained business impact,” says Said.
While there’s clear appetite from enterprises to buy into agentic AI tools, right now adoption remains in its infancy, Said adds. “This is due to both its newness, but also concerns around data privacy, cybersecurity, model hallucinations, regulation and the readiness of enterprise data for AI training.”
Startups building AI agents aren’t just competing with each other — they’re also competing with Big Tech.
OpenAI launched its AI agent product, Operator, to users at the start of the year and its release in Europe is expected to follow. Google, Microsoft and Amazon are all building similar products.
Despite the uncertainty around path to scale and routes to market, there’s still bullishness around the prospect of AI agents to generate big returns for VC. There’ll just likely be some big losers, too.
While Flaig says she expects the boom cycle of funding to continue throughout 2025, the risk of bust comes when startups that raised during the frothy times go back out to raise from 2026.
“By then, they will need to show proof points that they can deliver on their value proposition, exhibit acceptable levels of accuracy in workflows and show growing adoption within enterprise accounts,” she tells Sifted.
“For now, we haven’t seen many AI agent startups that can already prove this.”
