Smith Collection/Gado/Getty Images
AI startups promise to automate everything from your inbox to your Christmas shopping.
But talk to the people actually putting money into the sector and their restraint is revealing. Despite the flood of new tools, most investors stick to a small handful they find genuinely useful.
We asked AI investors which products they use day to day. The list was shorter than the hype would suggest.
Transcription products were around long before chatbots took off, and it is no wonder they are still a mainstay for investors. By one estimate, about half of office workers spend more than five hours a week in meetings, rising to more than 15 hours for the top 9%.
“One tool I can’t live without is Granola for note-taking,” said Payton Dobbs, a partner at Hoxton Ventures. “It helps keep me straight, summarizing the important points you need and leaving out the things you don’t.”
Founded in 2023, Granola plays in the same space as companies like Otter, which has been around for about a decade and boasts more than 25 million users. It lets users manually take notes while an AI assistant makes additions and reorganizes the notes as they go.
Inaki Berenguer, managing partner of LifeX Ventures, added: “At a high level it (Granola) is just another call transcription and summarization tool,” he said. “But in practice it’s far more accurate than anything else I’ve tried. It captures the nuance of conversations in a way that actually works.”
Large language models like ChatGPT and Anthropic’s Claude are now ubiquitous across the industry. That is no different for investors — but what’s revealing is how deliberately they describe using them.
When investors talk about AI agents, they tend to emphasize how limited they still are. “We use LLMs and associated tools and agents for doing deep research, alongside simple AI agents for automating tasks like web scraping and updates,” said Mikael Johnsson, co-founder of Oxx, a Stockholm- and London-based software investor.
Model choice, meanwhile, is far from interchangeable. Cecilia Ma, an investment manager at Norrsken VC, said it depends on how complex the task is. “From Claude and ChatGPT for general information to Specter and Harmonic for sourcing,” she said. “For deep research and complex tasks I leverage Claude or Strawberry.”
Specter and Harmonic are data platforms used by investors to find and keep track of early-stage companies, while Strawberry is a newer product aimed at more demanding research tasks.
And others use large language models for brainstorming. Lexi Novitske, a general partner at Africa-focused growth fund Norrsken22, said she uses Claude and ChatGPT “as a thought partner on new sectors.”
Beyond chatbots and research assistants, some investors are starting — cautiously — to experiment with AI agents, products designed to handle entire tasks on their own rather than responding to one prompt at a time.
“I’m still learning here myself, but I’ve started playing with Lua AI, an agentic platform to build a few agents that actually plug into my work,” said Novitske.
Founded in 2024, Lua AI is one of a growing crop of startups — along with firms like Cognosys and Adept — making software that lets AI carry out simple tasks across a company’s internal tools, rather than just generate text to answer questions.
“Some of our regular software tools are becoming a lot smarter themselves, but we're experimenting with agents to go through our own data and industry material and help a lot with pipeline building, pulling patterns out of what’s in our drive, and running quick scenario analysis on companies or markets,” Novitske said.
Even so, expectations remain measured. Research firm Gartner estimates that more than 40% of agentic AI projects will be canceled by the end of 2027 due to unclear value and rising costs.
They also come with risks. A recent survey of IT professionals found that among the 82% whose companies used agentic AI, many said those agents have gained access to unintended systems, used inappropriate data or even ordered things they shouldn’t have.
Despite companies’ best efforts, email remains an area where AI tools struggle to deliver.
A recent study by Stanford University found that 40% of workers had encountered AI “workslop” — machine-generated content that leaves colleagues frustrated, confused, and behind schedule. Each instance takes nearly two hours to resolve, carrying an invisible tax of about $186 per worker per month, the researchers calculated.
“I’ve played around with email AI agents like Fyxer, Superhuman and Shortwave, but never really found one I’m happy with,” said Dobbs, the partner at Hoxton Ventures. “I’m often reworking drafts so much that it’s easier for me to start from scratch.”
That hesitation fits a broader pattern. Investors still mostly trust AI to support their work in a limited way — summarizing, organizing, or scanning — rather than speaking or acting for them.
For all the sector’s promise, even the people funding it appear most comfortable where the stakes are low.
