Every new skill and component we add to our Tana Outliner systems raises the same question: are we improving our overall efficiency and effectiveness, or just giving ourselves more workflows to manage? Inspired by a recent viral post from OpenClaw creator Peter Steinberger, Mark walks through loop engineering: taking skills from previous Systems Labs and combining them into connected loops that can run with greater autonomy while keeping a record of every action in Tana Outliner along the way.
The idea is to move yourself back a layer. Agents get more room to operate, while Tana Outliner keeps the work visible, correctable, and grounded in the right context. The key distinction is that a real loop is stateful: each pass reads and updates a context database, so the next run behaves differently based on what changed. In these examples, that context database is Tana.
We'll focus on:
- Combining the routing and agent delegation skills into a single connected loop
- Deciding loop intervals and defining boundaries for what agents can do without approval, plus stopping conditions so a loop can't run away
- Saving receipts, decisions, and proposals in Tana Outliner so you can review the work asynchronously and just approve it