Although not everyone agrees on what an “AI agent” actually is, it remains popular. At a high-level level, these so-called “agents” promise to go several steps beyond mere chatbots, making decisions and taking actions on people's behalf. Some help with online shopping. Other companies may make their factories more efficient with varying degrees of autonomy.
That's why fledgling AI startup Across AI is quietly emerging to develop a “dynamic memory system” for complex enterprise workflows. And it's led by a founder who relatively recently sold his previous startup to IBM.
Cross AI targets chief revenue officers, sales teams, and more with a platform that connects to all internal and external corporate data sources. It then creates a shared “agent memory” that can be used to identify and evaluate new sales opportunities, identify risks, and suggest questions your sales team should ask customers.
“Sales teams often struggle to obtain and leverage the right information when they need it, including information about products, customers, competitors, and optimal processes,” said Co-Founder and CEO of Across AI. CEO Steven Mih (pictured above, center) told TechCrunch. . “Critical knowledge is often bottlenecked among a few experts or buried under vast amounts of unstructured data, leading to inefficiencies, slow decision-making, and costly errors. Existing AI solutions are unable to address this problem because they lack deep integration and situational understanding, do not have the ability to prioritize or adapt to new information, and treat all data equally. It happens often.”
Mih was previously co-founder and CEO of Ahana, backed by Google Ventures. The company built its commercial service on Presto, an open source SQL query engine spun out of Facebook in 2013. Mih sold Ahana to IBM for an undisclosed amount last year. And after 14 months at the tech giant, Mih took the leap in July to start working on his latest startup.
He collaborates with Dr. Niroufar Salehi (pictured above, left) and Dr. Afshin Nikzad (pictured above, right), distinguished professors at the University of California, Berkeley and Stanford University, respectively, to explore the effectiveness of AI. I've been researching ways to improve it. The system is in a “high-stakes” setting.
Across AI is still in its early stages and is privately refining the product with design partners. Aiming for a commercial launch in 2025, the company is currently raising a seed round of 575, co-led by Bobby Yazdani's Kota Capital and Village Global, a venture capital firm that includes Bill Gates, Mark Zuckerberg, and Jeff Bezos. Raised $1 million. , and among its supporters is Reid Hoffman.
Across AI Image credit: Across AI
Making memories
Across AI will be web apps and chatbots that connect to different parts of the enterprise stack (CRM systems, communication and collaboration tools, calendars, and everything else) to build memory and develop understanding of the situation. No matter where your users are working, you'll always be available to help them.
“Where your users are already, like Slack or [Microsoft] The team's app, the Across AI app, provides just-in-time assistance in the context of a user's existing workflow without interrupting the user's flow,” said Mih.
The company says this memory “continuously adapts,” retaining only what it thinks is relevant information and discarding old data. This very much depends on the context and requirements of the people using it, so the question arises as to how we can determine what is relevant.
Mih says this can be achieved through a “deep understanding of the workflow context.”
“The system actively tracks, timestamps and monitors information updates and recognizes when data becomes outdated or contradicted by new information,” he said. “Unlike traditional AI systems that treat all data equally, our agent memory system prioritizes information based on contextual importance. When possible, the app itself updates the inference If ambiguity exists, decisions are escalated to stakeholders such as sales managers and product managers.
Of course, businesses have been slow to adopt generative AI, as data privacy and security remain key concerns. The last thing a company wants is to have all its proprietary and sensitive data leaked to a third party and God knows what happens to it. As such, Mih says data security is a “fundamental aspect” of the startup's agent memory platform.
“Our memory systems operate within a secure in-house environment, maintain controlled access to sensitive information, and do not expose data to external models for training,” said Mih. “We plan to offer both SaaS and cloud-premises deployment options to meet enterprise security and compliance requirements.”
There's a subtle synergy between Mih's previous startup and his latest venture. Ahana focused on allowing users to query vast amounts of data through Presto, and Ahana handled all the complexities of setting up and maintaining the infrastructure. Across AI approaches the same problem through a different lens.
“I believe a key differentiator for AI application companies is their ability to enable users to quickly analyze large amounts of data, and at Ahana, that is exactly what we specialize in,” said Mih. Masu. “This experience has given me a deeper understanding of the challenges enterprises face in understanding complex data ecosystems that are often siled and difficult to navigate.”