Steven Mih’s first customers had found an advertisement in a newspaper. They were students looking for somewhere to live. The person answering their questions was a child, perhaps nine or ten, helping with his mother’s rental business. His other duties included cleaning apartments and mowing lawns. Before there were cloud marketplaces, there were actual lawns.
His mother had earned a real-estate broker’s license and entered the student-housing market. Mih remembered the telephone work particularly clearly: explaining which apartments were available to callers who had read the classified ads. The job put him between somebody’s question and the information needed to answer it. That position would become familiar.
Decades later, his working vocabulary includes distributed systems, query engines and AI agents. The businesses have changed considerably. Yet a useful thread runs through his career: making complicated technology useful to people who have something else to get done. He has approached that work as a salesperson, a hired chief executive and, eventually, a founder.
Eight managers, one rehearsal
Mih studied electrical engineering at UC San Diego, then went to Silicon Valley and found his way into sales. At AMD, new graduates entered a technical sales program that lasted a year. They rotated through product lines, learned the technology and practiced explaining it. There was time to learn before a quota arrived to make the introductions.
One detail gives the training its particular flavor: quarterly oral boards. A trainee conducted an hour-long mock meeting in front of eight managers and directors. Eight bosses is a substantial audience for a meeting that does not even produce a purchase order. The arrangement made preparation a practical necessity.
Years afterward, Mih credited the people who had invested in those graduates, including training leader Carol Harbin and the wider AMD organization. He remembered a culture of passing knowledge along. His account gives the apprenticeship more weight than a tidy origin story about an entrepreneur arriving fully formed.
His subsequent work included sales leadership at Cadence Design Systems, worldwide sales at Transitive and Couchbase, and chief executive roles at Aviatrix and Alluxio. At Transitive, he worked on sales and business development before IBM acquired the company in 2008. He had moved from established public companies into the less settled business of building a startup.
“I always wanted to start a company.”Steven Mih, 2021
A company of his own, built around an open engine
“I always wanted to start a company,” Mih said in 2021. Before that happened, he took the opportunity to run Aviatrix, joining the cloud-networking company when it had 18 employees. Being entrusted with somebody else’s young business gave him experience that would precede founding his own.
Ahana arrived in 2020. Mih and his co-founders built a business around Presto, the open-source SQL query engine. The premise was accessible even when the architecture required explanation: companies had data in many places, and people wanted answers without turning every new question into an integration project.
In a conversation with co-founder Dipti Borkar, Mih described SQL as a common language across that scattered landscape. Presto could help connect the questions to the repositories. Ahana offered help running the technology and commercial support to organizations that wanted its benefits without taking on all the operational work themselves.
That distinction mattered to his account of the company. Customers could use the open-source project themselves, or pay for assistance. He and Borkar were also involved with the Presto Foundation under the Linux Foundation. There was a product to sell, alongside a shared project that required a community to keep developing it.
By August 2021, Ahana had raised a $20 million Series A led by Third Point Ventures, with existing investors GV, Leslie Ventures and Lux Capital participating. The money was intended for engineering and product development as well as sales and marketing. A query engine still needed people to build it and people to explain why it belonged in a customer’s budget.
The meeting that could have been a video
Mih also made deliberate choices about how Ahana would work. The founders chose a fully remote organization from the beginning and drew on GitLab’s all-remote guidance. Communication required some design of its own. Putting everyone on a video call is easy enough; making that call necessary is a separate achievement.
He experimented with recorded video using mmhmm, the presentation tool later renamed Airtime. Short recordings gave colleagues something to consider before a discussion. His favored feature let him redo individual sections rather than rerecord an entire presentation. The appeal is understandable to anyone who has reached the final sentence and suddenly forgotten how sentences work.
“You have to remind yourself of the benefits of remote work,” he said. One benefit was concrete: he could more often take his children to school and collect them afterward. His account did not claim that remote work reduced the workload. It made room for a different arrangement of the day.
There is a related candor in his description of a sales forecast that failed. He once budgeted for a new distribution channel to supply a third of sales. The business did not arrive, and the second-half plan needed reworking. His lesson was patient execution over years. A spreadsheet may accept a distribution strategy immediately. Distributors retain the right to be less accommodating.
IBM buys the company. The questions continue.
IBM acquired Ahana in April 2023. The transaction also brought IBM into the Presto Foundation. For Mih, a company founded in 2020 had reached an acquisition by an enterprise technology business with a much longer history. The purchase price was undisclosed.
Ahana had raised $32 million by the time of the acquisition. Its work sat within a broader shift toward data lakehouse architectures, where organizations sought to combine flexible data storage with analytics. The acquisition connected its managed Presto offering to IBM’s enterprise business and extended IBM’s participation in open-source software.
Mih’s explanations of the technology often returned to flexibility. In a 2022 video conversation, he discussed open data formats and the ability to use different engines over the same data. The practical issue was what organizations could do with information after storing it. Storage alone did not provide an answer to a business question.
He was also explicit about community governance. In a late-2022 interview about SQL and lakehouses, he discussed an ecosystem whose separate components included storage, compute, catalogs and permissions. The resulting freedom created work for the people assembling it. Across these conversations, the sales executive remained occupied with the unglamorous distance between a promising architecture and a system someone could use.
A memory for the sales floor
In 2024, Mih co-founded Across AI with Nilou Salehi and Afshin Nikzad. It brought his experience in enterprise selling together with academic work on AI and algorithms. Salehi brought research from UC Berkeley; Nikzad brought expertise in algorithms and machine learning.
The initial target was business-to-business sales. Across set out to build memory for AI agents handling complicated workflows: information about a customer, changing priorities and details spread across enterprise systems. A system needed to recognize which information still mattered as circumstances changed. Retrieving an old answer was insufficient if the customer had moved on.
The company emerged with seed funding co-led by Village Global and Cota Capital. Its ambitions involved reasoning across internal and external information and providing verifiable insights and suggested actions. Mih was its co-founder and CEO at launch. Today, Across identifies Salehi as CEO and includes Mih among its advisors and investors.
The move into sales software had a certain symmetry. Someone trained to understand customer problems was working on software meant to help other sellers retain that understanding. The hard part was familiar: an organization might possess the information and still struggle to put it in the right conversation.
When software acts, who keeps the receipt?
Mih now leads Action State Group, whose work focuses on records of AI-agent actions. Its public materials identify him as founder and CEO. The company describes EvidenceBook as an open-source way to seal actions into a record that the owner keeps, with tools for somebody else to check it.
Consider the distinction through an ordinary transaction. Software may be authorized to place an order, attempt to place it, or receive confirmation that it succeeded. Those are different events. Treating them as interchangeable makes for a pleasantly brief report and a rather awkward reconciliation later.
Mih’s Agent Action Capsule proposal addresses that separation directly. The individual Internet-Draft records outcomes including execution, refusal, errors and timeouts, and distinguishes a dispatched attempt from an observed result. It remains work in progress, rather than an adopted IETF standard. Its scope is evidence about actions, including actions that were stopped.
The work has also taken him into public standards discussions and conference presentations. In July 2026, he confirmed abandonment of provisional patents associated with several disclosures. In September, he spoke at AGNTCon and MCPCon Europe in Amsterdam about independently verifiable action records.
Alongside the technology work, Mih serves on the board of San Francisco’s Francisco Park Conservancy. His current company’s biography places him in the city. It is a useful glimpse of a life with commitments beyond the next enterprise-software deployment.
From the rental telephone to open-source analytics and AI accountability, his work has kept returning to questions people need answered before they can proceed. An available apartment. A result from scattered data. A record another party can inspect. The scale has changed. The person at the other end still needs an answer that will hold up.