The company, which recently exited stealth, has raised $5.7 million in seed funding and is aiming to bring its first product to market by mid-2025.
Across AI, a startup developing the next-generation agentic AI software for improving enterprise workflows, has launched with $5.7 million in seed funding.
The San Francisco-based company is initially focused on developing its agentic AI memory technology for business-to-business sales processes and is aiming to bring its first product to market sometime in the second quarter of 2025.
The company was founded on the premise that current AI agent technology is unreliable for supporting complex business workflows. Today’s AI agents struggle to understand the nuanced context of – and relationships within – enterprise data, treat all data equally even if outdated, and lack the ability to prioritize, reason through, or adapt to new and changing information.
[Related: AI Companies With A Winning Hand: The 2024 CRN AI 100]
“The bigger picture is that business workflows haven’t been able to adopt AI very well. This is going to be the first enterprise sales AI, supported by what we call the next-gen agent memory,” said Across AI CEO and co-founder Steven Mih (center in photo) in an interview with CRN, joined by co-founders chief product officer Nilou Salehi (on left) and CTO Afshin Mikzad (on right).
The three started Across AI earlier this year. Across AI’s $5.7-million seed funding round was co-led by venture capital firms Village Global and Cota Capital.
The goal of Across AI’s product development is to create a dynamic AI agentic memory system with high-level reasoning that’s purpose built to support complex workflows.
“This is something that reasons over internal and external enterprise data and makes sure it provides accurate, verifiable insights and actions for the business workflows, starting with enterprise B2B sales,” Mih said. “Right now, agents are becoming a big deal and we think we’re positioned perfectly in that place.”
Mih is an entrepreneur who founded self-service data analytics company Ahana in 2020. Ahana was acquired by IBM in April 2023. Prior to that, Mih was CEO of data management company Alluxio and, before that, was CEO of cloud networking provider Aviatrix Systems.
CPO Salehi is a researcher in human-center AI and, as a professor at the University of California at Berkeley, has led ground-breaking research on AI system design including how AI can be integrated into complex workflows, according to the Across AI announcement. CTO Nikzad is a specialist in advanced algorithms and machine learning models who is currently on leave from his professorship at the University of Southern California.
The company’s agentic AI applications will be capable of reasoning deeply over unstructured data within each step of a workflow by using a proprietary combination of large language models and workflow-specific algorithms, according to the interview and a description of the company’s technology in the launch announcement.
Unlike early-generation knowledge retrieval systems, the Across AI technology will be able to continuously adapt to changing data, retaining only relevant information and discarding outdated data, and surface key insights based on contextual importance. Across AI’s development work is incorporating academic research from UC Berkeley and the National Science Foundation.
The system will provide accurate, verifiable insights and recommendations for business workflows. The company’s initial product will target B2B sales processes, which Salehi said can involve complex data – often from multiple sources – about customers, products, competitors and more.
“You need the agents to be able to remember what they’ve learned so far. “For instance, these are the customer’s pain points, these are their initiatives,” CPO Salehi said in the interview. “Building this memory is really, really hard because it has to understand concepts. And some of these concepts are very abstract and they’re always very different across different workflows. Building that and integrating it into the agent’s workflow is what we do.”
CTO Nikzad said the development work incorporates machine learning, AI and algorithm design techniques. Some AI-based algorithms will update the memory as new data flows in while others will support the AI agents that run on top of the memory and provide the insights and suggested actions for users, he said.
“The big vision is a memory for enterprise workflows and this memory is custom-built according to the workflow. That is what gives us an edge in terms of reliability,” Nikzad said.
Salehi said the Across AI product will also address issues of data access control and data governance that have hindered broader adoption of generative AI.
While the initial product will focus on sales processes Mih said the technology can be applied to a broad range of business workflows
Across AI is working with a number of early-adopter development partners – which Mih declined to identify – as it builds its product. Mih said they are large companies with many products, complex B2B sales processes, longer sales cycles, and thousands of sales representatives.
“They are the ones that want to transform their sales first. They’re the ones that want to get access to the agentic memory-backed Across AI solution,” Mih said.
