[ad_1]
Hthe Paris startup founded by Google alumni, made a splash last summer when, out of the blue, it announced a $220 million seed round before launching a single product. Three months later, with no product, that spot began to look like a catastrophic flood when three of the five founders left the company.”Operational and commercial disputes“.
But the company kept swimming, and today it’s announcing its first product: Runner H, an AI “agent” that targets businesses and developers across tasks like quality assurance and process automation. It is built on top of the startup’s “built-in” MBA based on just 2 billion parameters.
H has set up a waiting list for Runner H on his site. APIs will be released to those on the list over the coming days to use “off-the-shelf” agents previously designed by H, as well as to develop their own software, CEO Charles Cantor said. API access will also come with access to something called H-Studio to test and manage how your services work.
Initially, the use of these APIs will be free, after which a payment form will be offered.
Even with built-in MBAs, building and running AI isn’t cheap, especially as competition continues to raise money to develop its own products. TechCrunch also confirmed that H is raising a Series A, building what Charles Cantor, CEO and one of the co-founders remaining, describes H as part of the second era of AI — with LLM companies like OpenAI being part of the first era.
“We are fortunate that we are in a position to build our own models,” Cantor said. “But this second era will be as capital-intensive as the first.”
(Remember, the $230 million already raised — it appears to have added another $10 million since announcing it earlier this year — was a mix of equity and convertible debt. The long list of investors in that round included individuals like Eric Schmidt , Yuri Milner and Xavier Neil; VCs such as Accel and Creandum; and strategic backers such as Amazon, Samsung and UiPath.)
Cantor told TechCrunch that H had been quietly working with a few clients in areas like e-commerce, banking, insurance and outsourcing, who helped it improve the product.
“Everything (at H) is based not on our creativity but on customer feedback,” he said.
Runner H will initially focus on three specific use cases: robotic process automation (RPA), quality assurance, and business process outsourcing.
RPA is a field that has been around for years, and uses basic scripts to automate the most repetitive tasks that humans have to perform — like reading forms, checking boxes, and sending files from one place to another. In fact, not much robotic process automation was created using AI, even after AI began developing advanced skills. The idea with Runner H is that it will be able to run RPA across forms, sites, and other templates even when they are modified (something that may have broken previous scripts), and across a much wider range of sources.
Quality assurance can cover a wide range of applications, but Cantor said one of the most common applications by far is reducing the “maintenance burden” around website testing — checking page availability, simulating real user actions, or ensuring compatibility across payment methods — in Especially when no modifications are made.
BPO is a comprehensive field that will cover not only fixing and improving billing processes, but also accelerating how an agent uses and access data from different sources, and more.
There has been a race among established AI companies over the number of parameters that go into an MBA. (GPT 4 for example has 175 billion parameters.)
Runner H takes a completely different approach with only 2 billion parameters, both for LLM and computer vision-based “VLM.” Cantor’s argument is that this makes them significantly more efficient in terms of cost and operations, which is key when working to win and retain business deals, and the operational costs of H.
“We are specialists,” he said. “We are building for the age of agents.”
The company also claims it works: It says its combined models outperform Anthropic’s “Computer Usage” by 29% (based on WebVoyager benchmarks) as well as models from Mistral and Meta.
[ad_2]