A research paper is a peculiar sort of triumph. Years of work arrive as a PDF, colleagues applaud, citations accumulate, and the idea may never again encounter a civilian. Andy Konwinski has spent most of his career objecting to that last part. His preferred verb is not publish. It is ship.
At UC Berkeley, shipping meant releasing code beside the research. It meant helping build Apache Mesos, joining the team that created Apache Spark, and organizing rooms where users could learn what the software did. At Databricks, the company spun out of that work in 2013, it meant taking the early product to customers, building education and services, and creating Spark Summit, the gathering now called Data + AI Summit. Distribution was not decoration around the science. It was how the science acquired consequences.
Konwinski has since co-founded Perplexity, launched a venture fund for deeply technical founders, and committed $100 million to an institute for computer-science researchers. The nouns have changed. The plot has not. A promising idea is sitting in a lab. He wants to know what it would take to get the idea into somebody's hands.
The route he keeps rebuilding
The education of a question-asker
Konwinski was born in 1983 and raised in rural Wisconsin in a Jehovah's Witness family. His father was a machinist; his mother was a homemaker and school-bus driver. When he began questioning the faith as a teenager, the consequences were not abstract. He was expelled from the community at 18 and estranged from much of the world he knew.
A high-school counselor nudged him toward higher education. The route was wonderfully unglamorous: trade school, community college, then the University of Wisconsin-Madison. He earned a computer-science degree there in 2007 before crossing the country for graduate school at Berkeley.
Berkeley's AMPLab was less an ivory tower than a very clever garage with peer review. Konwinski contributed to Hadoop, co-created the cluster manager Mesos, worked at Google on the Omega scheduler, and became part of Spark's founding team. His 2012 PhD thesis concerned multi-agent cluster scheduling. The title was not destined for airport bookshops, but the work lived in the urgent territory beneath modern computing: who gets which resources, and when?
There was a social system around the technical systems. He helped organize Berkeley's Computer Science Graduate Entrepreneurs club and created AMP Camp, a hands-on workshop series. Spark Summit followed in 2013. Konwinski became a translator between researchers who knew what the code might do and users who knew what they needed done.
“Before you build anything meaningful, you need the space to ask questions, challenge assumptions, and follow your curiosity.”Andy Konwinski
Spark finds a company
Seven Berkeley researchers founded Databricks in 2013. Konwinski's job was not confined to a tidy executive rectangle. He worked on early marketing, helped customers deploy the software, built professional services and education, and later served as vice president of product for AI and machine learning. He also co-authored Learning Spark, a manual for the system he had helped carry into the world.
His Databricks co-founder Ali Ghodsi once captured Konwinski's particular radar: Andy would ask whether you had seen some fascinating new thing; six months later everyone would be talking about it, while Andy had already moved on. There is curiosity here, certainly, but also timing. Discovering an idea is one skill. Recognizing the moment when it needs a conference, a customer, or a company is another.
Konwinski stepped back from daily work at Databricks in 2019. He began investing alongside Andrew Krioukov, a friend from Wisconsin and Berkeley. Their fund, Computer Science Graduate Ventures, drew capital from computer scientists and invested back into researchers. One introduction led to Aravind Srinivas. Konwinski invested early, became a co-founder and founding president of Perplexity, and the researcher network supplied one of its founding engineers.
The episode became both a company and a prototype for a new fund. In 2024, Konwinski, Krioukov, and Pete Sonsini launched Laude Ventures with $150 million. More than 50 computer scientists joined the network behind it. The premise is specific: find technical founders close to the discovery stage, before the demo has learned to speak fluent venture capital.
A paper, a pub, and a much larger lab
Laude Institute began with an argument about agency. Computer scientists were spending plenty of time predicting what artificial intelligence might do to society. Konwinski and his former Berkeley mentor, computer architect David Patterson, wondered what would happen if researchers more deliberately chose the outcomes worth building toward.
The two sketched the question over Guinness at a Berkeley pub in early 2024. It grew into a paper, Shaping AI's Impact on Billions of Lives, written with Jeff Dean, John Hennessy, Finale Doshi-Velez, Mariano-Florentino Cuéllar, and other researchers. They spoke to people outside the usual technical circle, including scientists, policy leaders, and artists. Then came a salon, university talks, and eventually an institution.
Laude is pronounced “awed,” a small linguistic flourish in a field that often names things after acronyms. Its nonprofit institute launched in 2025 with Konwinski's $100 million anchor pledge and a board that included Patterson, Dean, and Joëlle Pineau. The institute offers fast Slingshot grants for projects that can move now, longer Moonshots support for sprawling public problems, and practical help for researchers trying to produce open-source tools or companies.
The arrangement is an answer to a structural tension in AI. University labs can publish openly but lack the resources of large corporate labs. Companies can supply compute and salaries but may keep their frontier work closed. Konwinski is trying to create a third lane: enough money and operating help for academics to build seriously, with the work remaining public enough to travel.
The moonshot acquires a clock
The first Moonshots call asked researchers to work in four areas: scientific discovery, frontline care, civic discourse, and workforce reskilling. The response was not modest. Teams submitted 125 full proposals involving 600 researchers across 47 institutions. A committee selected eight seed winners in April 2026, each receiving $250,000 and six months to develop a full plan for a $10 million multi-year lab.
The projects range from AI weather forecasts for regions poorly served by existing models to new tools for mathematical discovery, large-scale public deliberation, and skills for physical trades. Four runners-up and 13 honorable mentions also received funding, bringing the seed cohort to 25 teams and more than $4 million.
A moonshot without a clock is merely a handsome intention. Laude added a deadline. The eight teams present their work in October 2026; one is due to receive the larger lab. This is classic Konwinski engineering applied to an institution: define the interface, allocate scarce resources, insist on an artifact.
He has built other measuring devices too. The $1 million Konwinski Prize challenges AI systems to solve real software problems on a contamination-resistant version of SWE-bench. Its first round drew more than 600 teams and, usefully, did not produce a fairy-tale result. The best systems remained far from the 90 percent target. A benchmark that embarrasses the field can be more valuable than one that flatters it.
“Run at failure. I'm worried that if you optimize for your success, you're not going to pick a hard enough problem.”Andy Konwinski, Berkeley commencement, 2025
Success is a suspicious target
In May 2025, Konwinski returned to Berkeley's Greek Theatre, where he had received his doctorate, to address graduating students. He told them to stop succeeding. The provocation was not a billionaire's complaint about comfort. It was an engineer's warning about objective functions. Optimize a life for visible success and you may choose only problems with reliable applause.
“Think at the species level,” he told them. Think about a warming planet, jobs reshaped by AI, and the difficulty of speaking with people one disagrees with. The phrase could sound inflated in lesser hands. Konwinski gives it ballast with budgets, committees, milestones, and public code. Grand ambitions are tolerable if somebody eventually opens the repository.
There is a neat circularity to the current work. The student who found a home in a Berkeley lab now tries to preserve room for other researchers to ask difficult questions. The open-source organizer now funds artifacts. The product executive now teaches a PhD seminar called Research to Startup. The investor now wants academics to have the option not to disappear into a corporate lab before their ideas are ready.
Laude may produce companies, tools, and a few expensive failures. Konwinski appears comfortable with the assortment. He has already seen a university project become a piece of global computing infrastructure. The lesson he took was not that every paper conceals a Databricks. It was that nobody can tell which one does while it is still trapped inside the building.