Vince Trost is co-founder and CEO of Plastic Labs [1], a company that develops Honcho, an AI memory layer [1]. Based in the New York AI scene, Trost leads the organization in its work on memory infrastructure for artificial intelligence systems [1].
Founded
Insights & ideas
Vince Trost argues that AI agents should not be modeled on the persistent, identity-bearing nature of human agents, since large language models have no innate memory and are simply generating probable responses from whatever context is provided [1]. He contends agents are better understood as disposable, task-specific systems assembled just-in-time with only the relevant instructions, examples, and information needed for a given inference, then discarded [1]. He frames this approach as beneficial for efficiency, since resources can be provisioned on demand rather than idling, and as encouraging a modular, composable ecosystem where developers specialize in narrow tasks rather than competing with general-purpose models [1]. He also emphasizes privacy, noting that language models can make surprisingly deep inferences from small details, so withholding unnecessary context limits exposure [1]. Finally, he ties this design to AI safety, arguing that containing context to a single user and task makes failures easier to debug, and predicts disposable agents will spur a large expansion of economic activity [1].
Experience
- Co-FounderPlastic LabsFeb 2023 to Present
- Data Science EngineerPassage LabsApr 2022 to Mar 2023
- Research And Development EngineerPenn State UniversitySep 2018 to Apr 2022
- Kernel Block 4 FellowGitcoinSep 2021 to Nov 2021
Education
Penn State University · Bachelor of Science - BS, Data ScienceAug 2014 - May 2018
Media & appearances
- 'Big Brother' Star Vince Panaro Reveals Relationship Status ...Apr 7, 2025
Vince Trost, co-founder of Plastic Labs, discusses why current AI agents are flawed and proposes that agents should be built as disposable, task-specific systems assembled just-in-time with relevant context rather than persistent entities. He explains that large language models lack inherent memory and require developers to provide appropriate context through instructions, examples, and additional information, and argues this approach offers benefits for efficiency, privacy, and AI safety.
This page shows public professional information only, each fact cited. Is this you? Corrections or removal within 24 hours, no questions asked.