About Golemry

The technology that makes automation possible is not the same as the infrastructure you need to truly benefit from it. For agentic AI, Golemry is that infrastructure.

A decade of putting AI into production

As Chief Scientist at understand.ai (later acquired by dSPACE), I built human-in-the-loop annotation pipelines for autonomous driving, where models did the heavy lifting but a human had to review and correct the output before it could be trusted. Later, at Octomind, I built multi-agent systems that create and fix end-to-end browser tests, running unattended. The same problem showed up everywhere: the hard part is rarely getting a model to attempt the task, it is how to accelerate a human with it. That is where agentic AI fails today, too risky to leave alone and too tedious to babysit, so the review bottleneck is the problem worth solving.

Why Golemry

The idea clicked when I scheduled a routine task through an AI agent. Almost anything you can describe, you can now hand off: no workflow builder, no dedicated SaaS, just an agent and the connectors it needs. The catch is reliability and safety. An agent can leak a credential or go off the rails, and left alone it can silently break, drift from what you asked, or quietly stop running, with nothing watching.

Golemry is the answer I kept wanting. You describe the task, and it runs on safe infrastructure, with properly managed connectors and AI oversight that pulls you in only when something needs you. That is what makes it safe to hand an automation your inbox, payments, and data, and it closes the human-review bottleneck in a way I would trust with my own accounts.