Mihir Chintawar

Mihir Chintawar
Occupation	Co-founder and CTO of Random Labs
Known for	Co-founding Random Labs (Y Combinator S24)

Mihir Chintawar is an Indian-American software engineer and entrepreneur, known as the co-founder and chief technology officer (CTO) of Random Labs, a San Francisco-based startup building autonomous coding agents. The company was part of the Y Combinator Summer 2024 batch.

Early life and education

Chintawar is of Indian origin and is based in the San Francisco Bay Area. Prior to founding Random Labs, he developed technical interests in language model behavior, agent design, and what he has described as "expressive harnesses" — engineering frameworks in which the desired behavior of an AI model is shaped by the structure of the environment rather than by explicit instruction alone. He has written publicly on the topic of how models attend to context and how that attention degrades over long task horizons, themes that directly informed the technical architecture of his later work at Random Labs.

Career

Chintawar co-founded Random Labs in 2024 alongside Kiran Illindala. The company develops Slate, an open-source autonomous coding agent designed to work alongside software engineers on complex, long-running tasks. Slate is installed via the npm package manager as a command-line tool and is intended to operate within a developer's existing codebase.

A key technical feature of Slate is its built-in compaction algorithm, which the company describes as enabling the agent to manage its own context window dynamically. According to Random Labs, this mechanism surfaces the most relevant information at any given point during a task, allowing the agent to maintain focus over extended work sessions without degradation in output quality. Rather than employing a separate "plan mode," Slate is designed to implicitly plan by first researching a given task and then presenting its proposed approach to the user through an interactive discussion — an approach Chintawar has connected to his earlier thinking on expressive harnesses, where model behavior emerges from environmental structure rather than direct prompting.

In March 2026, Random Labs officially launched Slate V1, which was described in press coverage as the industry's first "swarm native" autonomous coding agent. The V1 release introduced a "hive mind" architecture capable of executing massively parallel engineering tasks across a codebase simultaneously. The system uses what the company calls a "dynamic pruning algorithm" to maintain context across large codebases. When a worker thread completes a task within this swarm architecture, it returns a compressed summary rather than a full transcript, enabling efficient coordination across parallel processes without overwhelming the central context. The company has positioned Slate as a collaborative assistant aimed at supporting what it calls the "next 20 million engineers," framing the tool as complementary to human developers rather than a replacement for them.

Chintawar's public technical writing has elaborated on the architectural decisions behind Slate. He has noted that large language models cannot attend uniformly across their context windows, meaning that information placed at different positions in a long context receives unequal weight during inference. This observation underpins the compaction and pruning strategies built into Slate, which are designed to counteract context degradation during extended autonomous task execution rather than relying on the model to handle long contexts natively.

Random Labs operates in the developer tools, open-source software, and artificial intelligence sectors. Slate has been used by engineering teams at companies including Gravity and Prava. The company's participation in Y Combinator's Summer 2024 cohort provided early institutional backing and positioned Random Labs within a network of developer-focused AI startups working on similar autonomous engineering problems.