I am rebuilding this site around the work that is actually becoming important.
For a while now I have been building hands-on agent infrastructure: memory, tools, approvals, dashboards, workflows, and enough glue to make agents do useful work instead of just answer questions.
Some of it worked. Some of it got overtaken by better tools. Some of it is still messy. That is fine. The useful part is the scar tissue.
The direction now is simple:
- release the pieces that are useful
- learn the enterprise agent platforms businesses are likely to buy
- stay model-agnostic where it matters
- apply the patterns to real business workflows
- teach what I learn as I go
I am less interested in AI hype than in workflow reality.
Where does the request enter?
What does the agent need to know?
Which system owns the truth?
Where does a human need to approve the next step?
What happens when something fails?
Those are the questions that matter if agents are going to do real work.
This site is now the home base for that work: learning logs, practical notes, and public releases when they are ready.