With May closed, three announcements from the period describe the same phase change for anyone working with agents.
The first concerns billing: separate monthly credits for agent builders, acknowledging that this use consumes differently from conversation and distorts the general plan.
The second concerns capability: a continuous execution mode, with the system working until the task is complete rather than answering question by question.
The third concerns measurement: a high score on a graphical interface operation test, which is the skill required for agents that use applications rather than write code.
All three point toward more autonomous use, and the counterpart is always the same and has to be decided in advance: what the agent can't do without confirmation, what its reach is inside the system, and what gets logged to allow review afterwards.
None of those three decisions depends on a supplier or a budget. All of them depend on someone writing the list before switching on automatic mode.