Can I see exactly what a run did?

Yes. Whether a run executed agentically while the workflow was still learning or ran on autopilot months later, it is recorded the same way and stays reviewable afterwards. Going autonomous changes how the work runs, never whether it is written down.

Open a past run and you get the ordered record: what triggered it, each step with its concrete inputs and outputs, every tool call, what the model was actually shown at the moment it decided something, how long each part took, where a human was asked and what they answered, what it cost, and any file the run produced.

You can also watch it happen. A run streams its steps, tool calls and reasoning summaries live, and that same stream is what you read back later. Every record is pinned to the exact version of the workflow that produced it, which is what makes "what did it do, when, with what data, and who approved it" answerable a year on.