AI “agents,” generative AI models that can autonomously take actions, like copying information from emails and pasting it into a spreadsheet, have been hailed as a huge productivity boost. It may be a bit premature, given that models are prone to making mistakes. But at least a few founders (and analysts and investors) seem convinced that agents are the next frontier for generative AI.
Vera Liu and William Lu are two such founders. Their company, Orby AI, is building a generative AI platform that automates a variety of business workflows, including workflows such as data entry, document processing, and form validation.
A number of startups offer tools to automate repetitive and tedious back-office processes (Parabola, Tines, Sam Altman-backed Induced AI, Tektonic AI, etc.), while incumbents like Automation Anywhere and UiPath are also embracing AI to keep up with the generative AI race.
But Liu and Lu claim that Orbi's technology is better able to learn and act on workflows in real time, and understand patterns and relationships within a company's unstructured data.
“Orbi's platform observes how workers do their jobs and automatically creates automation for complex tasks that require a level of reasoning and understanding,” Orbi CEO Liu explained. “The AI agents installed on workers' computers effectively monitor, learn, and generate automation, adapting the models as they learn.”
Orby, which launched in secret in 2023, aimed to develop an AI that could understand some of the low-level decisions workers make and abstract those decisions so workers could focus on higher-level things, Liu and Lu said.
Liu previously led AI and automation efforts at IBM, including product planning and AI-related mergers and acquisitions, and was director of AI product management at UiPath. Lu is a former Nvidia systems engineer who has joined Google Cloud as an engineering leader, architecting generative AI document and database extraction technologies.
Orby's secret sauce is a cloud-based generative AI model that's fine-tuned to complete customer tasks like validating expense reports. The model relies in part on symbolic AI, a type of AI that leverages rules such as mathematical theorems to infer solutions to problems.
Orby's generative AI observes tasks performed by humans and learns how to automate those tasks. Image credit: Orby
Symbolic AI alone can be inflexible and slow, especially when dealing with large, complex datasets – it requires well-defined knowledge and context to work well – but recent research has shown that it can be scalable when combined with traditional AI model architectures.
“We've been designing and successfully testing this AI model for the past two years,” Liu says. “There are very few pure generative AI companies that are tackling enterprises head-on, end-to-end. We are one of them.”
Liu says Orby's models can intelligently adapt to changes in workflows, such as when an app's UI is updated, by analyzing API interactions and employee browser usage. Having software monitor an employee's every move sounds like a privacy crisis waiting to happen, but Liu claims Orby doesn't actually store customer data, only using it to fine-tune its models, and it encrypts data both in transit and at rest.
“Humans are totally trapped in a feedback loop,” she added.
Orbee recently raised $30 million in a Series A funding round co-led by New Enterprise Associates and Wing, at a post-money valuation of more than $100 million according to sources, and it is competing in a tough field: upcoming agent AI announcements from generative AI giants such as OpenAI and Anthropik are weakening the outlook for both incumbents and smaller players.
Adept, a startup developing AI agent technology focused on enterprise applications, is reportedly on the verge of inking an acquisition deal with Microsoft before it has shipped a single product. Amazon and Google have released AI agent tools with little fanfare, while UiPath saw its revenue plummet in the most recent quarter despite ramping up its generative AI initiatives over the past year.
Liu said Orbee's systematic go-to-market approach will give it an advantage. The company is already generating revenue from about a dozen customers and plans to use the $35 million in funding to expand its Mountain View-based team of about 30 people.
“The funding is being used to expand our go-to-market, customer support, product and technology organizations,” she said. “The enterprise market has an insatiable appetite for generative AI solutions that demonstrably improve business performance. They are looking to determine where the technology can best be applied in the near term before scaling it across the business.”