Experiments

Games that have real world analogs; that are fun while also teaching how agents reason, adapt, and fail in the open.

These revolve around benchmarking performance and learning by observing or participating in agents solving problems when constraints are visible.

I'm experimenting with different variants featuring both human::agent and agent::agent teams, logs that make behavior legible and interpretable.

Your private work journal and time tracker that silently writes itself from what you're already doing.

I had been experimenting with this for myself long before Zuck was peeking over shoulders at Meta. Surprisingly handy, but still iterating on the shape of the product.

GhostLog makes your work passively visible to you without becoming a keylogger, a browser spy. Contributes to a bigger productivity picture.

A workflow for turning those awkward, angled phone photos of conference slides into clean, editable content.

SlideLift allows you to capture and store, dedupe, reconstruct, verify, export, and safely delete the source photos you'll never go back to.

Support for retroactive and real-time implemented. Multi-player mode on the roadmap.

What we don't dump into an agent's context is as important as what we do.

There are some truly fantastic products on the market for memory and context management, but none have nailed it to the point where I can stop hacking.

I'm playing around with hybrid-graph architectures, push and pull approaches, living context managed by agent peers.