← All speakers

Bio, Work & Ideas

Misha Kaletsky

Conference affiliation: Iterate · 2026

Misha Kaletsky is a software engineer at Iterate and creator of expect-type, an open-source library for testing TypeScript types at compile time. His projects make software interfaces more reliable, from command-line applications and PostgreSQL databases to experimental infrastructure for AI agents.

Kaletsky studied at the University of Oxford and worked at Microsoft, where his experience included C#, F#, TypeScript, Skype bots, and Microsoft’s Bot Framework. He subsequently worked in healthcare technology at Zocdoc, Haven, and Dr. B, building startup software primarily in TypeScript.

His open-source developer tools include trpc-cli, which converts tRPC routers into documented, type-safe command-line interfaces, and pgkit, which combines PostgreSQL clients, migrations, schema inspection, generated types, and administrative tools. He also created eslint-plugin-codegen and handy-redis.

  • Compile-time type testing: expect-type lets developers treat inferred types as behavior that can regress, covering generics, object properties, nullable values, and other difficult boundaries. Its assertions work within ordinary TypeScript workflows and integrate with Vitest; Kaletsky recommends concentrating these tests on reusable libraries and complicated edge cases.
  • Event-sourced agent harnesses: At AI Engineer Europe 2026, Kaletsky and Jonas Templestein explored an experimental architecture in which durable event streams record agent inputs, model outputs, errors, and state changes. Processors derive state through reducers while handling side effects separately. Kaletsky contributed an event-counting processor and a schema-matching helper for type-safe event payloads.
  • Compiler errors as agent feedback: Kaletsky proposed adding a TypeScript compilation step to JavaScript-based processors, recording compilation failures as events, and passing those events into automated repair workflows. The approach makes agent failures explicit, inspectable, and available for programmatic correction.

Read the topics behind these talks

1 conference talk

Key ideas

Scroll to read ↓

An append-only log, synchronous reducers, and explicit side effects provide the building blocks for a debuggable agent harness whose extensions can run on different machines.

  • What happened outside the agent’s log?
    0:16 ↗
  • Append Hello World, then watch it arrive
    5:18 ↗
  • Errors and controls are events too
    10:34 ↗
  • Turn a subscription into a processor
    15:57 ↗
  • Reconstruct state before performing side effects
    30:17 ↗
  • Derive the LLM request from state
    38:30 ↗
  • One persisted stream, many derived views
    46:50 ↗
  • Deploy behavior by appending JavaScript
    50:28 ↗
  • Error events and execution locations
    54:31 ↗
  • Let remote plugins contribute without blocking forever
    56:30 ↗
  • Who tracks delivery, and who may submit work?
    1:00:16 ↗

References