Skip to main content
Most requests finish in seconds. Some do not: reconciling hundreds of records, reading a long thread of documents, or working across several tools. Here is what Henry does when work is big.

​
Henry hands off to the background

When the remaining work will not fit in the current turn, Henry gives you what it has so far and continues in a background run. The background run has a much larger time budget, the full conversation, and the write approvals you already gave. It posts its result into the same thread, usually within a few minutes. In the web app you see Still working on this in the background under Henry’s reply until the result lands. In Slack or Teams the result simply arrives in the thread. Background continuation happens at most once per request and never skips approvals. If the background run needs a write you have not approved, it stages a confirmation like any other turn.

​
Footers you may see

Henry is honest about what it did and did not finish. These lines mean:
  • “This was a big one. I focused on the most important parts above. Want me to go deeper on any of it?” Henry finished the core of the job and trimmed lower-value detail. Ask for more and it continues.
  • “I ran out of time before finishing this one. Above is what I completed and verified so far. Tell me to keep going and I’ll pick up where I stopped.” The turn hit its ceiling. Reply “keep going” and Henry resumes from where it stopped.
  • “I ran out of time mid-run, so I’m already continuing this in the background. The finished result will post in this thread in a few minutes.” No action needed. Wait for the follow-up.
  • “…the partial work didn’t pass my final checks, so I’m not going to share something I can’t stand behind.” Henry chose not to post an answer it could not verify. Ask it to try again.
On scheduled tasks, a run that reaches its work budget ends with a scope note listing anything not done. Reply “finish the rest” in the thread and Henry runs it again immediately, or raise the task’s per-run budget in its settings.

​
Why there are limits

Every request has a time ceiling and a credit ceiling so one runaway job cannot stall the conversation or drain the workspace balance. Interactive turns wind down after a few minutes; scheduled runs get roughly three times longer. The credit ceilings are listed in Billing and credits.

​
Getting more depth

  • Ask for it. “Go deeper on the churn numbers” continues the work.
  • Use Standard mode rather than Quick for anything analytical.
  • For work you run repeatedly, make it a task. Tasks get a longer budget than a chat turn.
  • Henry can split research across parallel helpers when a job has many independent parts. It does this on its own for read-only work and verifies what comes back before answering.