AI business agent startup Bardeen pulls in strategic investment from Dropbox and HubSpot

Every day, employees across the world spend incalculable hours performing tediously repetitive tasks, such as converting documents into PDFs and then uploading them to a drive, from where they are sucked into a database and emailed to a team. While UiPath helped pioneer this kind of “robotic” process, it’s also pivoted to an AI-driven mode of operation, even as a number of startups (Signavio, Servicetrace and others) have arrived, snapping at its heels. Now a startup, Bardeen, is launching a service that automates such work for businesses on the heels of a new round of funding. 

Bardeen’s platform uses a natural language interface to automate repetitive knowledge work. The company has secured $3 million in this new round, taking its total funding raised to $22 million. That might be only vaguely interesting, were it not for the fact that the investors who participated in this round will provide significant distribution for the platform. Both Dropbox and HubSpot have become strategic investors in the startup via their venture investment arms (Dropbox Ventures and HubSpot Ventures), and the two companies will also help distribute Bardeen’s tech, releasing on Thursday. 

If you think “Zapier with more AI” then you’d get close to what Bardeen offers, but its product is a bit more sophisticated that what that may suggest. It’s very much built as a platform for the average person inside a business to perform repetitive tasks, rather than something built for IT departments, and a demonstration of Bardeen’s interface to TechCrunch showed how easy it was to automate complex workflows.

The platform seems capable of quite a lot: It can copy-paste text from one document to another, research related information on the web, and put all that info into an email and send it. The startup says it has more than 300,000 users and over 1,000 paying customers, including Deel, Miro, Kearney, WPP and 10Web.

Founded in 2020 by Artem Harutyunyan and Pascal Weinberger, Bardeen’s agent platform runs as a browser extension and is context-aware, so that the agents can conduct a “planning” step after receiving instructions from the user — which, Bardeen says, helps with repeatability. It also says its assistant continually learns from usage patterns.

It also integrates with 100 tools like Microsoft 365, CRMs and sales platforms.

That repeatability feature is important because getting an AI platform to give you the same answer twice is not easy. The lack of that predictability in a business setting will kill any product stone dead.

As CEO Pascal Weinberger told TechCrunch: “The problem with other AI solutions is that they just couldn’t achieve repeatability. You would give it the same task to repeat and it would do two different things. It’s, by nature, how these language models work, but makes it pretty hard to use for real business applications.”

So what is Bardeen’s approach?

“You input your prompt, such as take the meeting notes, turn it into a PDF, extract email addresses and send the PDF to each person, for instance,” said Weinberger.

“The platform runs it through a language model, and this is where the differentiation begins. It has a planning stage. So the model figures out that it has to go to the calendar and extract the calendar event, extract the email addresses, create a PDF, etc. It can do that, and then I can just type, ‘Also send a PDF to Pascal on Slack,’ for example,” he added.

Once the model has figured out a plan, it will stick to it: “So the next time I ask it to do the same thing, [the process] has become a learned skill in the same way you would teach an assistant or junior. So I can do the whole thing purely by writing in natural language. Everyone can build automation like this.”

Of course, as usual, the question is, what LLMs does the platform utilize?

Weinberger says they use Gemini to translate questions and OpenAI GPT models for “specific automation exercises.” He added, “But every week, there’s a new model that comes out and we have a benchmark to see which tasks a model is better at.”

While incumbents like Zapier, UiPath and others are racing to keep up, it looks like Bardeen is poised to leap ahead, at least for now.