The company.

I build in layers. Some products carry no AI at all. They just need to work, and doing that well is its own discipline. In others, AI does the unglamorous part: capturing information a person would otherwise miss, and synthesizing it faster than they could by hand. Useful, but still incremental.

The layer I care about most is different. It asks whether AI lets you do something that was never possible before, something that changes the shape of the product instead of its speed.

I look for that same opening in software: not how to do the old thing faster, but whether the old thing needs to work this way at all.

I also build with local models, sized to the smallest one that will actually run the job on someone's typical machine, for tasks that need real privacy or can't carry a cost per use, like reading a folder of screenshots or restructuring files. Models keep improving, and some of what I build is aimed at where the model will be in a year, not where it is today. Whether open-weight models running locally end up mattering to that future, I don't know yet. Further out is physical AI, where the category itself is still being invented.

Different problems call for different layers. Matching the depth of the tool to the depth of the problem is the actual skill.

We're always looking for good ideas. If something in your week keeps costing you time or money and nobody has built the fix, write in and tell us about it. The writing, the research and everything still in progress live at byzyb.ai.

If something we made isn't doing what it promised, the address below reaches a person.

General & support

rh@pklabs.ca

Founder

byzyb.ai