-
Generative AI startup Orby has raised $35 million to automate business workflows using large-scale action models (LAMs) instead of the more common large-scale language models (LLMs).
-
Orby’s platform automates tasks by observing user behavior, eliminating the need for users to write code, fitting into the “no-code” trend.
-
The platform also uses generative process automation (GPA), which is different from traditional robotic process automation (RPA).
Orby AI, a generative artificial intelligence (AI) startup that has raised $35 million in funding, enters the industry with a vision to automate businesses with little to no effort. Notably, their AI approach uses large-scale action models (LAMs) rather than the large-scale language models (LLMs) used by most AI platforms today.
Here’s how it works:
The Orby platform lets you build automations while monitoring the behavior of all your apps, workflows, and basically everything you do. Orby looks at the HTML, mouse and trackpad positions, associated images, and the types of repetitive tasks you’re performing. The platform then uses code to automatically generate each workflow.
Orby CEO and founder Bella Liu says the platform fits into the emergence of the “no-code” enterprise technology category, which is essentially about making tools more accessible to business users. But the problem is that most no-code tools “are not really no-code,” Liu told Fierce Network.
(Orby CEO Bella Liu, via Orby)
“A lot of tools in this generation claim to be ‘no code,’ but when business users actually start using them, they need to know how to write software. They need to understand all the logic, variables, arguments,” she added.
Orby’s platform doesn’t require users to write any code or do anything out of the ordinary – the system simply watches, learns, and builds the automation code itself.
LAM approach
The key difference here is that Orby uses LAM. Many AI models to date are language-based, requiring users to understand how to formulate queries to extract information and insights. Popular generative AI (GenAI) applications that use language-based models include OpenAI’s GPT-3, Google’s Smart Compose, and IBM Watson Assistant.
LAMs are relatively new, and are action-based; they don’t rely on prompts, but rather rely on observing and learning from the user’s actions as they “move” along with them. They provide cognitive assistance to the user in a way that LLMs and queries do not.
According to one analyst, Shelly Kramer of theCUBE Research, the beauty of the Orby platform is that LAM can “monitor, learn, automate, and adapt.” Krammer noted that Orby’s solution works alongside users, like the Grammarly app. Just as Grammarly watches how users compose their writing and applies AI to improve it, Orby watches everything users do and learns their movements, behaviors, preferences, workflows, and more. It’s essentially “auto-running.”
“This is a larger, more robust model of action than a typical LLM, and one with a very compelling use case for organizations looking to adopt and infuse gen AI across their entire organization,” she told Fierce.
Kramer said Orby isn’t the first AI platform to leverage LAM, but the approach isn’t necessarily common yet. LAM will “definitely” be the next big trend in AI, he added. “As you know, this is a competitive space, and we think a lot of the use cases for LAM are compelling both from a development standpoint and from a user standpoint.”
GPA is a juice “worth squeezing”
Orby’s work is also part of a market category called generative process automation (GPA), which is distinct from traditional robotic process automation (RPA).
RPA uses rules-based software bots to automate simple, repetitive tasks like data entry. GPA uses AI to handle more complex tasks that require understanding and creating content, like composing emails or making decisions. That’s why Orby calls its platform “rules-free” AI.
Liu said RPA deployments take time, often taking months or even years for a single process. Additionally, RPA projects are costly because they require specialized developers, data scientists and IT support, and some customers feel it’s “not worth it.”
Keeping humans in the loop
Realistically, it’s important to note that many of today’s AI use cases still require some human oversight, and Orby is no exception.
For example, if changes are needed after the automation has begun, a human can always make the changes.
Because of AI’s ability to learn through experience, the platform may encounter situations it has never seen before, Liu said. In cases where there is “low confidence,” Orby will send confirmation notifications to users and learn from their feedback.
“That’s actually one of the things we really believe in: constantly monitoring humans,” Liu said. “Current AI technology is actually very good. It can make the workflow very accurate from the get-go just by observing the user once, but it’s usually very hard for AI to be perfect, because then even humans aren’t perfect.”
Future competition
Kramer said there doesn’t appear to be any “major competition” for Obie right now, but there may still be some players operating under the radar.
“Just as Orby has remained in stealth mode for the past year or so, they are certainly not alone,” Kramer added, “I think it won’t be long before we see more conversations about LAM and more activity and entrants in the space.”
For example, Microsoft and Salesforce are both considering using LAM, and she said they’re “certainly not alone.”
To date, Orby has raised a total of $35 million in funding. In June, the company announced a $30 million Series A funding round it had raised last year. Why the delay? “We’ve been very busy with product and customers,” Liu says. “Now we’re really ready to announce our Series A and all the traction we’ve achieved to date to the world.”
