Building
Agentic Workflows
Personal and client-facing workflows that use AI agents as amplifiers for judgment, not replacements for it.
The shape of the work
I build agent workflows for the parts of work that are easy to drop when the calendar gets loud: the handoff, the review loop, the forgotten decision, the source material that should have stayed close.
The point is not autonomy as spectacle. The point is continuity. A good agentic system should help a person return to context with less friction and more honesty about what is known, what is guessed, and what needs judgment.
How I tend to build
I start with the operating principles before the tool. What should never be lost? What should be checked every time? Where does the human need more room, not less? Then I build small enough that the system can be trusted before it becomes impressive.
The boundary
I do not want agent work to become a hiding place from harder product or relationship decisions. The best versions serve the spine of the work: they preserve context, sharpen review, make handoffs more honest, and leave the human with a clearer next move.