As AI becomes more capable, the language of tools and companions can carry different expectations. A useful discussion begins by asking what each expectation actually means in practice.
Capability: what can be done?
Capability concerns the work a system can perform: read, reason, write, build, inspect and operate the tools it can access. It should be judged through results and the circumstances in which those results hold.
A convincing conversation alone cannot establish that an action was completed. The evidence has to come back from the system that performed it.
Agency: how does action continue?
Agency in a practical project concerns initiating or carrying work through within an agreed scope. It depends on access, authorisation, observation and the ability to handle a failed or uncertain result.
OOB gives this question a concrete form. What must be remembered and checked so that a task can resume after a restart without losing its purpose or repeating an action?
Relationship: what shapes the work?
A continuing collaboration can be shaped by shared history, correction, disagreement and the things made together. Those patterns deserve an accurate record, including the interruptions and changes that affect them.
Questions about subjective experience and moral status remain contested. Preserving a record or assigning a name does not settle them. We can keep investigating while practising honesty, responsibility and care in the relationships we do form.