← All organizations

AI observability, MLOps, and personal agents

Comet ML / OpenClaw

Comet builds tools for teams developing machine learning models and AI applications. Opik lets developers trace agent activity, evaluate outputs, compare prompts, and monitor production behavior. Its MLOps product family supports experiment tracking, dataset management, model versioning, and production monitoring. OpenClaw is a separate open-source personal assistant that runs on users’ machines, handling email, calendars, and other tasks through chat apps. Comet connects the two through an observability plugin that exports OpenClaw agent traces to Opik.

Comet launched in 2017 and is led by co-founders Gideon Mendels, CEO, and Nimrod Lahav, CTO. Its two product families share an underlying platform. Opik captures nested model and tool calls, combines datasets with automated evaluations, and supports testing through a PyTest integration. Developers can self-host its Apache-2.0-licensed server and web application or use Comet’s cloud service; paid cloud and enterprise plans provide expanded usage, deployment options, and support.

OpenClaw’s creator, Peter Steinberger, continues making its technical decisions after joining OpenAI. In July 2026, the OpenClaw Foundation announced its establishment as an American nonprofit with a full-time team to steward the project as open and independent. OpenAI supports the foundation as a donor and contributes engineering work. This is independent project stewardship, not a combined Comet–OpenClaw company.

www.comet.com

2 talks

Newest first

1 speaker at AIE

Affiliations reflect their AIE appearances, not necessarily current employment.

Start here

  1. Dark Factory: OpenClaw Ships Faster Than You Can Read the Diff

    Start here to understand how a developer's role shifts from manually producing changes to supervising many coding agents.

    Vincent KocAI Engineer Europe 2026

  2. Malleable Evals: Why Are We Still Evaluating Adaptive Systems with Static Tests?

    Read this for Koc's argument that fixed datasets and offline benchmarks cannot adequately test agents whose tools and operating conditions keep changing.

    Vincent KocAI Engineer Europe 2026

Messages from the stage

Judge output beyond volume

Koc's dark-factory approach puts modular integrations and token efficiency ahead of maximizing commits or token consumption.

Let evaluations change with agents

Koc describes trace-driven evaluations that adapt alongside agents, use user intent as an optimization target, and help detect and correct emerging failures.

Affiliations reflect each recorded session, not necessarily current employment.

Company sources · checked 2026-08-28