Cybersecurity, AI agents, and networking.
What would it take to give AI agents more freedom to act, without losing sight of what they can do, what they did, and whether it worked?
Explore the projects01 / About
WYŁOM Research is an independent research practice based in Cambridge, working across cybersecurity, AI agents, and networking. Founded and run by David Godlewski.
We explore how increasingly autonomous software fits into systems people can still understand and control. That brings together questions usually treated separately: how software is organised, how it communicates, and how it can fail or be misused.
02 / Research
From the architecture of agent autonomy to software running on unexpected hardware.
[01] Architecture research
AOS explores how an operating system could give agents explicit authority, versioned state, and recoverable changes. It is architecture research, not a working OS: the aim is to test the foundations before committing to an implementation.
Read the AOS research[02] Agent interfaces
Prompt64 turns the Commodore 64 Ultimate into a terminal for monitoring and controlling AI coding agents. A native client and host bridge bring live activity, approval prompts, and code changes to a deliberately constrained interface.
Explore Prompt64[03] Experimental networking
tailscale-vita is a from-scratch Rust implementation of a Tailscale client for the PlayStation Vita. It brings encrypted networking and direct peer connections to homebrew software, working within the handheld's platform constraints.
Explore tailscale-vita03 / Approach
Some questions need an architecture before they need code. Others are best explored by building something and seeing where it breaks. We move between the two—from thinking through how agents should operate to making software run on hardware it was never designed for.