On March 27, 2026, I gave a lab share session on how I use AI agents in day-to-day life and work.
The session was less about generic “AI productivity” and more about a shift in how I think about software and daily workflows. What interests me now is not just asking a model for help in one isolated moment, but building systems that carry context, keep working after I stop typing, and start to feel like part of the environment itself.
The main point of the talk was simple: the interesting change is not that chatbots are getting better. It is that agents can now be organized into workflows, connected to tools, and given memory and context that persist across tasks.
I used Claude Code as the main example because it has been the tool that made this shift concrete for me. What matters most is not just code generation, but the ability to run commands, inspect outputs, call tools, and iterate inside a real working loop.
I also shared two systems that reflect where my thinking is going. RyanHub is my attempt to build personal AI around shared context rather than isolated features. AgentOS is my attempt to treat agents less like chats and more like an organization, with roles, memory, routing, and review.
The broader mindset shift, at least for me, is from executor to leader, from isolated tools to systems, and from individual sessions to institutions. Once agents can accumulate memory and operate within stable workflows, the question stops being “what prompt should I use?” and becomes “what kind of system am I building?”
If there was a single theme across the whole session, it was that I am less interested in AI as a smarter interface and more interested in AI as infrastructure: something that can coordinate work, carry context, and become part of the environment in which research and daily life happen.
If you want the full deck, it is available here: