Why AI Agents Need
Progress Reporting
Long-running autonomous work should not disappear between assignment and outcome.
An AI agent can do useful work without being easy to supervise. It may read files, call tools, inspect APIs, change plans, retry steps, and wait on external systems. If the only visible states are "started" and "done," the operator has very little to work with.
Progress reporting turns agent work from a black box into an observable process. It shows what the agent is trying, what it has completed, what evidence it found, and where it is blocked.
Autonomy needs checkpoints
The more freedom an agent has, the more important checkpoints become. A person should be able to see that the agent accepted the task, loaded the right context, called the expected tools, reached a milestone, or stopped because confidence was too low.
Progress is not just a spinner
A spinner says the system is busy. Progress reporting says what kind of busy. For agent workflows, useful progress may include tool calls, records examined, documents summarised, errors recovered, decisions deferred, and actions awaiting approval.
It improves intervention
Operators do not always need to wait for failure. If an agent is stuck retrying the same tool, reading the wrong source, or skipping expected inputs, progress reporting gives someone a chance to intervene before the final result is wrong or late.
Agents are operational processes
OpenTrace treats agent workflows like other background work: they need progress, notes, metrics, payloads, durations, and milestones. The goal is not to expose every token. The goal is to make meaningful work visible while it is still happening.