Rust NYC x OpenAI: Safer 'unsafe' & Barnum: The agentic workflow engine.

PLEASE NOTE: FOR SECURITY PURPOSES, YOU MUST FILL OUT THE OPENAI SIGN UP PAGE AS WELL AS THIS, USE THIS LINK: Rust NYC x OpenAI - the password is Rust@OAI2026

Join us on Tuesday, April 28 at OpenAI NYC. Doors open at 6:15 p.m. to give attendees plenty of time to grab food, drinks and socialize, and the talk begin at a slightly earlier start time of 7:05pm.

Robert Balicki, Staff Engineer at Pinterest
Barnum: The missing workflow engine for agentic workflows

Robert Balicki is a staff engineer at Pinterest, where he works on the Web Platform team and helps the company adopt GraphQL. He previously worked on the relay team at Meta. He organizes the Rust NYC meetup, and his side projects include Isograph, an opinionated, compiler-driven framework for building data-driven apps ([https://isograph.dev](https://url.uk.m.mimecastprotect.com/s/-dPfC28QZI8qn2VUnfXS5B1L8?domain=isograph.dev)), and Barnum, the missing workflow engine for agentic workflows ([https://barnum-circus.github.io](https://url.uk.m.mimecastprotect.com/s/pL3OC36P8T21LgXhqhJSQIMYc?domain=barnum-circus.github.io)).
Robert will be talking about Barnum, the missing workflow engine for agentic workflows. LLMs are incredibly powerful tools, and we're tasking them with increasingly complicated workflows, from writing code to controlling our computers. However, if the assigned task is too complicated, the LLMs become forgetful and cannot be relied upon to precisely execute the workflow. Plan mode helps, but it's also inherently difficult to express a complicated, tree-like, conditional workflow in a linear markdown file.
This is where Barnum comes in. The key insight of Barnum is that you can't build reliable tools upon a fundamentally unreliable primitive. So, instead of hoping that an inherently probabilistic agent can advance a deterministic state machine, we have the workflow invoke the agent. Developers provide Barnum with a detailed workflow, which the engine then drives. Tasks that are deterministic (such as type checking) are not sent to agents. Agents are asked to do just the minimal tasks where they are the best tool for the job (such as reading files and reasoning about them). This means it's structurally impossible for agents to wriggle out of requirements or cut corners, and because these invocations are ephemeral, there's no problem with context rot.
Don't want to bother writing a workflow? Don't worry, the APIs are fully statically typed and so agents are incredibly good at doing that part too.
Barnum has already been used successfully to generate hundreds of clean-up PRs, automatically performing complex migrations, remove dead code due to fully shipped experiments, power a RAG search engine and validate each claim in documentation. Turns out, precisely specifying the work up front increases the reliability of agents, allowing you to go much further than if you just YOLO'd it!

Predrag Gruevski, Member of Technical Staff at OpenAI
Performance Wins at OpenAi scale: Safer 'unsafe' with Codex and Miri

Predrag (https://predr.ag) is a performance engineer best known for his work on the cargo-semver-checks linter, the Trustfall query engine, and his blog post about debugging WiFi that only worked while it was raining. At OpenAI, Predrag works on distributed storage systems that power the training of frontier models.

At OpenAI scale, every millisecond counts. Rust is phenomenal for this out of the box, but occasionally `unsafe` is necessary to squeeze out the last drops of performance from a system. This talk is a case study of how we used unsafe Rust, Codex, and miri to ship a tricky optimization in a massive distributed system. We went from idea to production in record time — since we used Rust, everything worked on the first try!

Lawrence Harvey is Rust NYC's official recruitment partner, with Ross providing support as a co-organizer and financial support.

The space is generously sponsored by our partner OpenAI, who will have a number of their engineers from the Rust heavy teams present for you to mingle with and learn more about the Rust based projects they're currently building!