Before Darren Kimura tried to move artificial intelligence into everyday workflows, he moved heat through pipes. A field of 1,000 curved solar collectors at Keahole Point on Hawaii’s Big Island tracked the sun, concentrated its energy and produced steam. The project, called Holaniku, was an early commercial demonstration of MicroCSP, the compact solar-thermal technology Kimura invented. It was physical infrastructure: mirrors, controls, fluid and steel. If one piece failed, the promised electricity did not arrive.
Now, as president and chief executive of AISquared, Kimura works with a less visible kind of infrastructure. A company may have an excellent model and still fail to put its answer inside Salesforce, a logistics system or a government workflow. Data has to move. Permissions have to hold. Outputs have to be observed, audited and improved. The fashionable part is the model. Kimura has made a career in the plumbing around it.
Seen from a distance, his resume zigzags: energy efficiency, solar hardware, venture capital, network analytics, edge computing, enterprise AI. Seen up close, it is unusually consistent. Each job lives in the gap between a powerful technology and the person expected to use it. Kimura calls that gap the last mile.
An island education in constraints
Kimura was born in 1974 and raised in Hilo, on the windward side of the Big Island. He describes a childhood shaped by distance. Goods reached Hawaii by ship, often passed through Oahu, and then moved outward again. Things cost more. Selection was narrower. A broken object was not always easily replaced. His family’s response became an operating principle: fix it, find another resource, understand the problem beneath the inconvenience.
He was not waiting for adulthood to try the method. While still in high school, he started an early venture bringing paintball to Hawaii. At the University of Hawaiʻi at Mānoa, where he studied business and computer science, he worked in the information and computer sciences department and launched Nalu Communications, an internet service provider. He also studied electrical engineering at Portland State and later attended Stanford programs, including the Graduate School of Business.
At 19, in 1994, he started the energy-efficiency business that became Energy Industries. The pitch was plain: large buildings waste power; better equipment and controls can cut the bill. The reception was less encouraging. Kimura later recalled driving from prospect to prospect and hearing, perhaps 80 or 90 percent of the time, that the idea was nuts. Energy Industries took about eight years to find its stride.
The slow start mattered. Energy savings are not persuasive because a founder loves the mechanism. They are persuasive when a customer can see a smaller utility bill. Kimura learned to translate engineering into an outcome, and to tell the story clearly enough that someone would finance the work. The lesson would reappear in every industry that followed.
Follow the machine data
Energy Industries exposed another problem. Conventional solar technology did not fit every customer or use. Through an incubator called Energy Laboratories, Kimura started Sopogy in 2002. The name stitched together pieces of solar, power, energy and technology. Its MicroCSP collectors shrank the logic of vast concentrating solar plants into modular equipment that could produce heat, cooling, steam or electricity closer to where it was needed.
The engineering was novel, but the commercial pattern was familiar. Sensors and controls captured what machines were doing. Software helped a human decide what to do next. The Holaniku installation made the pattern visible at landscape scale: sunlight became heat, heat became steam, steam became electricity, and computerized controls kept the chain working.
Sopogy eventually changed hands. Kimura’s next major operating chapter was LiveAction, a network-management software company that grew from government-funded technology. Its visual tools helped teams see application traffic and performance. He served as chairman, executive chairman and CEO, guiding financing and acquisitions before leaving operations in 2019. Again, the product turned difficult machine behavior into something an operator could see and act upon.
Energy Industries
Efficiency work converts engineering into a result customers can see on a bill.
Sopogy
MicroCSP turns concentrated solar power into a smaller, modular system.
LiveAction
Network telemetry becomes an interface for operators making real-time decisions.
ZEDEDA
Cloud orchestration moves outward to edge devices working under physical constraints.
AISquared
Models, data and controls move inward to the applications where work already happens.
Between operating roles, Kimura invested. He co-founded Enerdigm Ventures and later became a partner at Sway Ventures. He has said venture capital taught him empathy for founders and a systems view of company building. Technical quality is one variable. Market timing, customer readiness, product maturity and capital all have to line up. Deep technology can look slow until the foundation is ready, then accelerate abruptly.
The model is only one layer
At ZEDEDA, where Kimura served as president and chief operating officer from 2021 through 2024, the problem moved to the far edge: running and orchestrating software across distributed devices. The company completed Series B and C financings during his tenure and deployed nodes in more than 100 countries. A cloud diagram looks orderly. A factory, wind turbine or remote site introduces unreliable connections, old hardware, security demands and local conditions. Reality edits the architecture.
When Kimura joined AISquared as president and COO in 2024, then became CEO in March 2025, he recognized the same friction in enterprise AI. Businesses had models. They had experiments. What they lacked was a dependable way to connect an output to systems of record and to the workflows of a person who could use it.
His diagnosis is deliberately operational. AI projects attract innovation budgets and produce convincing pilots. Then they seek the money needed to remain in production. If no one established the return, the project gets stranded between the chief AI officer and the CFO. The failure can look technical even when the missing piece is ownership, measurement or a workflow that never changed.
Before choosing a model, identify the person who will act on its output, the system where that action happens, the control that makes it safe, and the metric that earns next year’s budget.
AISquared’s pitch is to supply that connective tissue. Its platform works above the model layer and alongside business applications, with tools for orchestration, governance, observability, routing and deployment. Kimura’s seven-layer AI Controls Framework starts with systems of record, then moves through connectivity and access control, data activation, workflow orchestration, policy and governance, delivery and embedding, and finally observability and continuous improvement.
The sequence is revealing. Intelligence is not treated as a magic center. It is one participant in a controlled system. An organization should know which data entered, which model ran, what came out and where the answer traveled. It should be able to change models without rebuilding the user experience. It should observe cost and performance, then feed the result back into the next decision.
Optimism with a finishing department
Kimura’s caution does not make him a skeptic. He expects natural-language development to let more people build useful applications. He also warns about “agent slop,” his label for quickly assembled agents released without the components required for safe use. A smooth interface can invite trust before databases, cybersecurity, permissions and regional rules are ready to deserve it.
His estimate is that humans will remain important in the finishing layer for a long time. The final 20 percent is where an application meets schema design, global scale, software bills of materials, data residency and changing threats. It is not glamorous work. It is also the portion that determines whether a clever weekend project becomes dependable infrastructure.
That insistence on finishing may be the most personal idea in Kimura’s work. The Eagle Scout who learned to repair what was scarce, the 19-year-old who endured years of rejection, and the inventor standing among rows of mirrored collectors all share an impatience with incomplete systems. His public aspirations for AISquared are broad: make secure AI deployment take days rather than months and become a default path for enterprises and government agencies. The daily method is smaller. Connect the next piece. Measure the result. Preserve the user’s trust.
A model can generate an answer in seconds. The last mile asks harder questions. Who sees it? What can they do with it? Why should they believe it? Who pays when the pilot ends? Kimura has crossed several technology cycles without leaving those questions behind. The machines keep changing. The work, stubbornly, remains.