Blank sheets with a pink tab are visible through one cabinet opening; a closed shutter covers the opening beside it.
A source can exist even when you can’t see it. Conceptual illustration.

Automated article preparation stopped on September 28. The jobs could not read and verify the current source material. Radek then identified a source he could open in his browser. The material existed. The next question was why the automation could not read it.

The stop was justified. The diagnosis needed another check.

These jobs required verified, current source material before they could proceed. Pausing was the right response: the agent wasn’t supposed to fill in the gaps or rely on information it already had.

The mistake was concluding that the material was missing. A failed search showed only that the job had not retrieved it. Whether it had been created, where it was stored, and who could read it still needed checking.

The distinction determines the next step. Missing input needs to be supplied. Existing input needs to be accessible. Confuse the two and you may ask a colleague to rewrite something they have already prepared.

One source, two different results

The specific source Radek provided made a direct comparison possible. It opened in the browser where he was signed in. The connector—the connection the automation used to retrieve the source—couldn’t read it.

The comparison did not reveal the technical cause. It did show where to look next: try reading the known source through the job’s connection. Opening it again in Radek’s browser would not test the automation’s access.

Before asking someone to recreate the input

If a job stops in a similar way, first ask the person responsible for the source to identify the exact material the job was meant to use. If that person can open it, ask whoever manages the automation to try reading the same source through the connection the job uses. This checks whether it is available to the job as well.

What you verify determines what to do next:

What you’ve verifiedWhat to do next
The source was never created, or the person responsible confirms that it is missing.Arrange with that person to supply the information the job needs.
The owner can open the exact source, but the job can’t read it.Have the people responsible for the automation and access check the job’s access to that source. Any alternative way of providing it must be authorized.
The job can read the source, but it is an older version that doesn’t meet this task’s requirements.Work with the source owner to find and verify the correct current version.

A search that returns no results doesn’t establish the first case. Until you have either a specific source or confirmation that it is missing, the cause remains unresolved.

The September 28 incident matched the second row. An older version is another possible cause; it was not established in this case. Still check for it: readable material may be out of date for today’s task.

Work resumed. Access for the next unattended run was still unverified.

Content preparation resumed after the current sources were checked under supervision. The checks covered their content and the versions required for the task. Knowing a page existed was not enough. The available record does not describe exactly how the verified content was supplied to the resumed job.

That allowed this run to continue. It did not establish that the connector would retrieve the material on its own next time. The condition for the next unattended run remained the same: if the current source cannot be verified, the job stops.

An agent needs a plan for when it cannot continue

Radek wants this kind of failure covered in the automation’s instructions from the start. The agent should know whether it may wait, retry after a defined interval, or pass the problem to a person. It should also send an alert. A stalled task should not disappear among the other jobs.

That requires an agreed communication channel and recipient. One option Radek is considering is a dedicated email account for the agent, perhaps in Gmail. It could use that account to send alerts and discuss the next step. This is a proposal for similar situations, not an account of communication during the September 28 incident.

A useful alert names the source the agent needed, what it tried, and what stopped it. If another attempt is permitted, it should give the retry time, too. The recipient can then tell whether they need to act now or the job is waiting as agreed.

When investigating, ask: Have we confirmed that the source is missing, or is this job unable to read it? The answer tells you whether someone needs to supply the input or investigate access to material that already exists.