Oct 15, 2026
Why some reflections take longer
This article describes the response process now available as See the story (beta).
While building Reflection Partner, I tried several ways of getting a response to stay faithful to a person’s experience. A reply could sound fluent and helpful while missing a qualification, flattening a tension, or giving one part of a message attention at the expense of another. Improving the instructions helped, but I kept finding situations where the process itself needed to change.
The difficulty was that an early reading could determine the whole response. But working through a message can reveal something that was not apparent at first: a relationship between two concerns, a detail that changes the meaning, or a reason to look again at something said earlier.
Reflection Partner now has a process that can revise its plan as that work develops. For a more involved reflection, it can return to a part that needs attention, draw on relevant earlier conversation, and reconsider how the pieces belong in the reply. In technical terms, this is a bounded agentic loop. It has limits on what it can do and how long it can continue.
I use coding agents in my own work. They make it familiar that a useful result may require investigation and revision before it is ready. Human experience has its own complexity. Several feelings can coexist; a small qualification can change what a sentence means. The fact that a message is written in ordinary language does not make that work simple.
The extra work is in service of fidelity: attending to the experience without losing what makes it yours. It does not give the app greater authority over your life, or permission to interpret more than you invited. A careful response can still be short.
The cost is waiting. Responses are not instant, and more involved reflections can take a couple of minutes. You can leave the app and choose to be notified when the response is ready. That makes the wait easier to accommodate, but it does not remove it.
A longer process can still miss something. I’m continuing to tune the system so that additional work earns its place, and simpler reflections can take a shorter path. The goal is a faithful, useful response, with as little waiting as that work requires.