Benjamin Harvey asked for fifteen minutes. He got breakfast at the Four Seasons. In February 2022, the AI Squared founder sent Roger Ferguson, the former chief executive of TIAA, a LinkedIn message about the software he was building for financial services. It was a small request directed at someone with a large calendar. Ferguson invited him to meet in Washington. By 2024, Ferguson was an investor and was joining the company’s board.
There is something pleasingly ordinary about the mechanism. No secret handshake, no elaborate introduction ceremony. A message, an invitation, a conversation. The relationship took time to grow, but its beginning fitted inside the same application where most people congratulate former colleagues on promotions.
Harvey had already committed considerably more than a morning to his idea. With his wife’s agreement, he put $500,000 from retirement savings into AI Squared and exhausted a $20,000 credit card limit. Those figures describe a personal wager, not a financing strategy to imitate. They also give the breakfast some weight. He had something to show, and quite a lot riding on whether anyone else would see its usefulness.
The useful thing was an answer to an unglamorous question. Once an organization has built an AI model, how does the person doing the work actually get to use it? Harvey had spent years meeting that question in different forms. The ambition was large. The obstruction was often a piece of ordinary software.
First, get the computer working
Harvey grew up in Jacksonville, Florida, in a family with seven children. His father was a pastor and an AT&T project manager who brought broken computers home. For a child curious about machines, broken was an invitation. Harvey learned to take them apart and repair them. A computer that worked meant games could run again; the reward for understanding the machinery arrived on a floppy disk.
His older brother supplied another kind of access, introducing him to calculus while Harvey was in sixth grade and later sharing computer science fundamentals. Technical knowledge entered his life through family, before it came with an institutional badge. That detail matters in a career that would eventually involve several of the country’s most guarded and credentialed workplaces.
He went to Mississippi Valley State University on a football scholarship and studied computer science. He then earned a master’s degree at Bowie State University in 2011 and a Doctor of Science in computer science there in 2015. His personal academic site records a cryptologic computer science certificate through the University of Maryland, Baltimore County, as well.
The route ran through historically Black universities. These were the places where his early curiosity became formal training, rather than a decorative footnote to a later startup. By the time investors entered the picture, he had already spent years learning how to make difficult systems yield something usable. Repairing a machine and explaining a model are different jobs, but both require attention to what happens after the clever part is finished.
Bachelor’s degree
Master’s degree
Doctor of Science
The work that stayed on the shelf
Harvey joined the NSA in 2009 through its Cryptologic Computer Science Development Program and completed that program in 2012. His career there included data science leadership and the position of chief of operations data science. He has also described leading data analysis of the national security impact of the Edward Snowden leaks. These were environments where an answer’s reliability and destination mattered together.
His account of the deployment problem is unusually concrete: a team of roughly forty data scientists, with about ninety percent of their work failing to reach a mission application. That is his description of his experience, not a census of every AI project. The models could function. Getting their output into the applications people actually used was the obstacle.
An organization can have talented scientists, substantial data and a working prediction, then still leave the intended user waiting. The result lives somewhere else. The user’s task continues inside an established application. Connecting the two requires more work, more coordination and more time. A successful experiment can become a very well-qualified spectator.
Harvey’s brothers served in the Army, including deployments to Afghanistan and Qatar. Their service made the consequences of inaccessible or delayed information personal. When he later explained the origin of AI Squared, he resisted the convenient invention story: “There was no true 'aha' moment”. Frustration accumulated. The gap persisted. Eventually, it became a business problem he wanted to spend his own time solving.
“There was no true 'aha' moment”Benjamin Harvey, 2023
A familiar screen, a different answer
In 2019, Harvey left the NSA for Databricks, working as a solutions architect. He encountered the commercial version of the same difficulty. Companies wanted value from data and machine learning, but the path from a technical result to a daily decision could be awkward. The audience changed; the missing connection remained.
AI Squared’s answer was to put the intelligence inside existing applications. Consider a financial adviser looking at a customer list. A model elsewhere might rank which prospects are most likely to become customers. The adviser benefits when those rankings appear beside the list already on screen. Asking someone to visit another system, move information around and reconcile the results adds friction at precisely the point where the prediction should help.
At FinovateSpring in 2023, Harvey presented with chief technology officer Michelle Bonat and chief operating officer Alvin McClerkin. The company’s demonstration described a goal of compressing AI integration from eight months to eight hours. Treat that as the demonstration’s claim, rather than a stopwatch guarantee for every installation. The point of the presentation was the distance between building an AI capability and making it available.
By the autumn Finovate event, AI Squared was demonstrating generative AI features, with Bonat and marketing consultant Andre Llewellyn presenting. The forms of AI were changing quickly. Harvey’s original concern remained legible: an enterprise already owns applications, data and working habits. Software has a better chance of being used when it respects all three.
Capital for the connection
In June 2022, AI Squared announced a $6 million seed round led by NEA, with Ridgeline participating. On April 17, 2024, it announced $13.8 million in Series A funding led by Ansa Capital, with Latimer Ventures and existing investors NEA and Ridgeline taking part. Together, those announced rounds total $19.8 million. They are company financing, not Harvey’s personal wealth.
Ansa’s investment case emphasized deployment, feedback and iteration. AI Squared could connect models to business applications and collect feedback that helped improve model performance and interface design. The investment thesis gave commercial shape to the irritation Harvey had described from government work: organizations were investing in AI and still needed help putting it to work.
Allan Jean-Baptiste, Ansa’s co-founder and general partner, joined the board alongside Ferguson. The company’s early public customer lists included Vanguard and the U.S. Air Force. These relationships placed the platform in settings where adoption involves more than persuading somebody to try a new app. Existing systems, security expectations and procurement processes all affect what can actually be installed.
Harvey’s public pitch for the round included a blunt observation: “Far too many companies aren't getting enough ROI from AI.” It is a less romantic fundraising line than a promise to reinvent intelligence. It also meets a buyer where the buyer lives. An organization can admire a model’s capabilities and still ask whether its employees can use the result before the next meeting.
Two rounds. $19.8 million combined.
Making room for other people
A founder who began by repairing computers has to learn when to stop being the person repairing everything. Harvey has described the move from engineer to CEO as a shift toward building the company that makes the product. That means trusting other engineers and spending time on customers, resources and relationships. Technical competence helps him understand the work; delegation lets the work grow beyond his own hands.
Mentorship appears throughout his career. Matei Zaharia, a Databricks co-founder, encouraged the AI Squared idea. At Maxar, where Harvey led data science work involving satellite imagery, executive Tony Frazier was another mentor. These connections brought different kinds of experience into view: research, software development and the practical business of serving large institutions.
Harvey has also put himself on the other side of that exchange. In 2022, he and AI Squared colleague Lloyd Pierre mentored three Bowie State teams participating in the NSA University Challenge Showcase. Their projects ranged from geospatial machine learning to secure communications and similarity aggregation. The students were working on named problems, with people who could connect that work to professional settings.
His advice to aspiring data scientists favors a portfolio showing applied experience across frameworks. A hiring manager should be able to see what someone has tried to do. That preference fits his own history: knowledge becomes persuasive when it has somewhere to go. For a student, the destination may be a first job. For a founder, it may be a fifteen-minute conversation that turns into breakfast.
The next question is permission
Harvey’s position at AI Squared has changed. Darren Kimura joined as president and chief operating officer in October 2024; by March 2025, company material identified Kimura as president and CEO. Harvey remains the company’s founder, with the chief executive chapter part of his career rather than his current title.
On February 26, 2026, Harvey announced Archetypal AI and its flagship platform, GOVERN. The announced focus was translating policies into enforceable rules for AI systems. The announcement described three beta-testing entities and a planned commercial deployment in the third quarter of 2026. That schedule was a stated plan, not evidence that the deployment occurred.
The new venture shifts attention from access to authority. Once an AI system is present in a workflow, what is it allowed to do? Who sets the boundaries, and how are those boundaries enforced? Harvey’s move connects two practical questions about the life of a model outside the laboratory. Getting it onto the screen creates another obligation: deciding how it should behave there.
His career has travelled from a broken computer in a Jacksonville home to systems built for organizations with complicated rules. The settings grew larger. The underlying demand stayed recognizable. A useful machine must work for the person depending on it. Harvey keeps returning to the space between technical possibility and actual use, where the demonstrations end and somebody still needs an answer.