Andy Konwinski

Career

Konwinski completed a PhD in computer science at UC Berkeley, where Konwinski contributed to Hadoop, co-created the Mesos cluster manager and was part of the team that created Spark. During graduate school Konwinski also worked at Google on the Omega cluster scheduler. Konwinski then co-founded Databricks, serving as VP Product AI/ML, creating the Data + AI Summit and co-writing the O'Reilly book Learning Spark.¹⁴

AWS's startup profile of Perplexity names Konwinski as one of its four founding engineers in 2022, and Konwinski's site says Konwinski was the company's founding President. Konwinski later co-founded Laude Institute, which gives grants and support to computer scientists turning research into startups or open-source infrastructure, and Laude Ventures. Konwinski also teaches a Berkeley PhD seminar called Research to Startups.⁶¹²⁷

Agents and benchmarks

In May 2024 Konwinski introduced Headless Terminal (ht), an open-source program built with the creator of asciinema that wraps a terminal so that language-model agents can send it input and read text screenshots of its state. In December 2024 Konwinski announced the K Prize, a competition run at Laude for open-source AI that can solve a contamination-free version of the SWE-bench coding benchmark, with a new test set collected after the submission deadline.⁸⁹¹⁰¹

At Laude, Konwinski's team co-created the Terminal-Bench benchmark and the Harbor framework with Ludwig Schmidt's group at Stanford. The Terminal-Bench paper, posted in January 2026 with Konwinski among the authors, presents Terminal-Bench 2.0, a set of hard tasks in computer terminal environments for evaluating AI agents.¹⁵²

Work and contributions

• Terminal-Bench Published research

  Co-author of "Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces" (2026). Konwinski's site says Laude co-created Terminal-Bench and the Harbor framework with Ludwig Schmidt's group at Stanford, and Laude's site lists Konwinski on the Terminal-Bench team.⁵¹²

• Mesos Published research

  Co-author of "Mesos: A Platform for Fine-Grained Resource Sharing in the Data Center" (NSDI 2011); Konwinski's site says Konwinski co-created Mesos during graduate school.⁴¹

• Apache Spark Open-source code

  Listed as a committer on the Apache Spark committers page. Konwinski's site says Konwinski was part of the team that created Spark and co-authored the O'Reilly book Learning Spark. The apache/spark repository is Apache-2.0.³¹¹¹

Timeline

1. Mar 2011Mesos, co-authored by Konwinski, was published at the 8th USENIX NSDI symposium.⁴¹²
2. 2013The Omega cluster scheduler paper from Konwinski's work at Google received the EuroSys Best Student Paper award.¹
3. 2022Co-founded Perplexity with Aravind Srinivas, Denis Yarats and Johnny Ho.⁶
4. May 31, 2024Introduced Headless Terminal, an open-source tool that lets agents operate terminal programs.⁸⁹
5. Dec 10, 2024Announced Laude Ventures on Konwinski's site.⁷
6. Dec 12, 2024Announced the K Prize, a contamination-free SWE-bench competition for open-source AI.¹⁰
7. Jan 2026Co-authored the Terminal-Bench paper on benchmarking agents in command-line interfaces.⁵

Other documented roles

Roles the sources document that are not confirmed as current.

• Co-founder and founding President, Perplexity¹⁶
• Co-founder; VP Product AI/ML, Databricks¹

Honors and recognition

• NSDI Test of Time Award (Mesos) · USENIX, 2021¹²¹
• EuroSys Best Student Paper (Omega) · EuroSys, 2013¹
• EuroSys Test of Time Award (Omega) · EuroSys, 2023¹

Where they work

• Laude Institute and Laude Ventures 1 person in the wiki
• UC Berkeley 27 people in the wiki

Related people

• Jaime SevillaEvaluation and benchmarks · Datasets and data
• Jerry LiuAgents · Datasets and data
• Jeff HuberDatasets and data · Evaluation and benchmarks
• Greg KamradtEvaluation and benchmarks · Agents
• Sasha LuccioniEvaluation and benchmarks · Datasets and data
• Kilian LieretAgents · Evaluation and benchmarks
• Simon SuoDatasets and data · Agents
• Alexander ToshevAgents · Datasets and data
• Eugene YanEvaluation and benchmarks · Datasets and data
• Rohan TaoriEvaluation and benchmarks · Datasets and data