About Jason
I started working in IT at fourteen.
A summer internship became a systems job while I was still in high school. I have changed roles and companies many times since then, but I still learn by opening the system and finding out what it does.
I have spent most of my career working on software, cloud, storage, and distributed systems. The largest chapter was twelve and a half years at Microsoft, including eight years on Azure-side storage work supporting OpenAI and Microsoft AI. After Microsoft, I spent part of 2025 and 2026 at VAST Data working on inference data architecture and economics.
I also write code, direct coding agents, and operate Kubernetes at home. My adult son, James, and I build Verdify, a greenhouse control system where AI may tune bounded parameters but deterministic firmware owns safety. Those projects keep me close to the code, hardware, and failures.
I did not take a traditional path. I grew up in Boulder County with two parents who never finished high school, very little money, a lot of instability, and plenty of love. A teenage IT job and the early web gave me a way into the work. I learned by getting stuck, asking questions, and staying with a problem until I understood enough to make something useful.
A few adults kept showing up
I was born in Boulder in 1981 and have lived in Boulder County for almost all of my life. My mother cleaned houses. The father who raised me drove trash trucks for his entire career. Neither of them graduated from high school. They divorced when I was nine, and after that we moved through trailers and rentals across the Front Range. By the time I graduated, I had attended seven schools.
The summer before eighth grade, my mother lost her job and we were about to be evicted. Her brother called Social Services. I was taken into the custody of the State of Colorado and lived in a group home while a court fight played out. Eventually, my dad got custody.
I was not alone. In fourth grade, I joined one of the earliest Boulder County classes of the “I Have a Dream” Foundation. Its coordinators stayed with the same group of kids for years. When our phone had been disconnected, one of them would drive to wherever I was living and knock on the door to make sure I was okay. I remember that more clearly than almost anything the program paid for.
The program also helped me get a summer internship in the University of Colorado’s financial aid IT department. I was fourteen. That internship turned into my first real technical job: Assistant Network Administrator, administering AIX, Windows NT, and Novell systems and building small Perl and JavaScript tools while I was still in high school. At eighteen, I enrolled in evening classes at Front Range Community College, paid my own way, and worked full-time. After roughly three semesters, I chose the job. I do not have a completed degree.
The web gave me somewhere to learn
In March 1996, when I was fourteen, Centennial Middle School’s online newspaper, Vocal Point, published an article I wrote about Bosnia-Herzegovina. I wrote other pieces and helped publish the paper. That same year, I built a ThinkQuest astronomy site called Shooting for the Stars with two other students. What stayed with me was the feeling of making something that strangers could reach.
I volunteered as a student Internet Guide for the school district, then started asking my way through Perl, IIS, and Linux on public mailing lists. The questions were not elegant. I was trying to route forms, update files, compile kernels, and make redirects work. Later the subjects changed to networking, PHP, MySQL, VPNs, S3, virtualization, and storage, but the method stayed the same: try it, inspect what broke, ask a specific question, and try again.
Side projects were where I tried the next thing
I started Brio Solutions in my early twenties and built LAMP-stack sites and hosting systems beside my full-time job. I wrote tools to backtest stock strategies, experimented with an eBay auto-bidder, and launched a small social-news site called NewsPicker. I built a local-news-to-RSS bridge because the newspaper did not have a feed and a request-echo tool because I needed to see exactly what a web service was sending.
In 2007, I wrote a Gmail IMAP tool to extract, deduplicate, and rank email addresses, then split the results into files small enough for social-network contact imports. That utility became GmailExtract, then MineMyMail, a bootstrapped product that people were still using in 2016. None of these projects became a large company. They solved problems I had and forced me to learn the systems around them.
Today I operate a private platform for coding agents on Kubernetes. James and I also build Verdify around a real greenhouse. The model may propose a change, but the ESP32 owns safety and telemetry has to show whether the change worked. That boundary is intentional.
Support, software, delivery, architecture, and product
My day jobs moved from support into software, project delivery, architecture, and product. At SHL, I went from client support to software development and technical integration. At NewsGator, I delivered enterprise projects and later worked as a solution architect on SharePoint social collaboration. At Avanade, I led delivery and directly managed engineers and project managers. At Neudesic, I worked on product management for Pulse, its enterprise social product.
I joined Microsoft in April 2013 as a SharePoint Premier Field Engineer. By late 2014, I was a Cloud Solution Architect helping partners and customers move to Azure. The field organization was still small enough that I helped interview many of the people joining it.
In 2017, I moved into Azure Storage product management. By the end of that chapter, my public-facing title was Group Product Manager, Azure Blob Storage. I led product managers and technical program managers working with large storage customers and on the Blob experience itself: APIs, SDKs, client tools, tiering, pricing, and usability. I led through product ownership and dotted-line accountability across engineering, not direct engineering management.
The constraint was simple: do not leave GPUs waiting on storage
My first meeting with OpenAI in 2017 changed the scale of the storage problem in front of me. Training used expensive GPU clusters, and storage could not be what left them idle. The useful measures were sustained throughput, checkpoint completion, data locality, and recovery. A generic latency number did not describe the problem.
We designed checkpointing around that constraint. Memory and local or peer NVMe handled the fast path. Azure Blob provided durable recovery without forcing every checkpoint to wait on object storage before training could continue. The team built the capability; my role was to turn the workload into product requirements, architecture, capacity, and an operating plan.
The namespace had to grow past one storage cluster. I co-architected Azure Scaled Accounts so one name could span multiple clusters. That required placement, routing, and rebalancing as capacity was added. Data also had to be near the GPUs before compute came online, so we used regional storage and service-side copies over Azure’s private network.
I later led Azure Blob’s storage and data workstream for Project Fairwater. I also owned the Blob developer surface and the operating loop around these workloads: capacity reviews, telemetry, regression detection, incidents, root-cause analysis, and the roadmap changes that followed.
OpenAI’s 2019 report Dota 2 with Large Scale Deep Reinforcement Learning names me under Microsoft Azure Support. At SNIA’s Storage Developer Conference in 2024, Jegan Devaraju and I explained Scaled Accounts, Blobfuse, hierarchical namespace, checkpoint movement, and global data movement. My work was infrastructure and product underneath the models, not model research.
I left Microsoft in October 2025 after about twelve and a half years and joined VAST Data as VP of Cloud Product Management. During that shorter chapter, I worked on inference behavior, KV-cache capacity, SSD endurance, throughput, hardware life, and the cost of keeping AI systems running continuously. My time at VAST ended in 2026.
Architecture is partly a translation job
A design is not useful if the mechanism disappears when it leaves the engineering room. I like sitting with an engineer inside a distributed-systems failure, then carrying the useful part into a product decision, customer conversation, or public talk. The details have to remain accurate enough for someone to act on them.
I have done that at Microsoft Azure IaaS Day, SNIA’s Storage Developer Conference, VAST Forward, and theCUBE. In 2025, the Federal Trade Commission played my recorded Microsoft Azure deposition in federal court in FTC v. Meta. I explained how public cloud let startups use large infrastructure without buying the hardware and building the organization first.
I also care about what technical systems change outside technical rooms. My December 2025 Longmont Museum talk asked what happens when human labor and talent are no longer the main limit on economic output. I later brought versions of that conversation to other groups in Longmont. Those discussions were less about model benchmarks and more about energy, work, governance, and what people do when a machine can perform more of the work that once defined them.
I returned to the program that had kept showing up for me
I returned to “I Have a Dream” first as a volunteer, then a mentor, sponsor, donor, board member, secretary, and eventually Board Chair. I am now a former Board Chair.
I knew the work from the other side. A coordinator had once knocked on my door when our phone was disconnected; years later, I had a chance to help the institution keep showing up for other kids. My family and I still sponsor a Dreamer.
At twenty-one, I also ran for Longmont City Council in Ward III. I received 1,331 votes, or 28.24 percent, against an incumbent. I lost. The race taught me how different public responsibility feels from arguing about policy from the sidelines. More recently, I have spoken with local groups about AI, work, energy, and the decisions communities will have to make for themselves.
Heritage is not the same thing as relationship
I learned young that the dad raising me was not my biological father; I had been conceived through a sperm donor. Much later, DNA and family-history research identified my biological father. I keep his name and his family private because they declined contact. The public-safe truth is still striking: he had been a decorated fighter pilot who flew hundreds of combat missions in Vietnam, then became an orthopedic surgeon and volunteered his skills for children. Learning about his life answered some questions about where parts of me came from.
That is heritage, not relationship. The father who raised me gave me a different inheritance: grit, work, and a complicated example of what showing up means. I have never missed one of my son’s band concerts. That matters more to me than anything a DNA test can explain.
DNA also brought something joyful. I found a half-sister, Mandy, through 23andMe, and our reunion became a national 23andMe television and customer story in 2016. It was a real connection where none had existed before.
My adult son, James, builds Verdify with me. Emily and I also have two younger daughters. Their private lives do not need to become content for my family to be central to this story.
What is all of this for?
I was a vocal atheist in my twenties, an agnostic through much of my thirties, and at forty-four I find myself more religious than I have ever been. I went to church on Christmas Eve in 2024 for the first time in about thirty years. My faith is not conventional or settled. I wonder whether the intelligence we are building changes how we understand creation, agency, and our own role as creators.
I think automation can create material abundance. That does not tell us how to live together, distribute power, or help people find meaning when work no longer tells them who they are. I worry more about powerful systems in human hands than about a mathematical system suddenly deciding it hates us.
I am also rooted in Boulder County. My family now lives in a historic Longmont home where Emperor Akihito and Empress Michiko stayed for three days in 1994. I have filled it with solar, batteries, cameras, climate controls, a greenhouse, and more Home Assistant entities than any house reasonably needs.
It is a house, not a lab. But it gives me real systems to operate, real failures to diagnose, and projects I can build with my family.
Where I am now
I still work from the mechanism outward.
That has been the pattern since the first web page: inspect the unfamiliar system, learn it well enough to be useful, and explain what I found. These days that includes persistent coding agents, local models, greenhouse controls, home automation, and whatever failure is open in front of me.
The work I want is still the same: hard systems problems, visible consequences, and people willing to say what works and what does not.