HACH developer: slow AI agents need instant ack, not silence

The piece is written from the operator's side of the desk rather than the model's. Its starting observation is that a colleague who needs time says "let me check" before going quiet, while an agent that processes silently leaves the user staring at a read receipt. That gap, the developer argues, is what triggers the familiar behaviour of asking the same question again by email five minutes later.

The three patterns build on each other. The immediate acknowledgement exists to close that gap, but it is deliberately kept free of any promise about the outcome, so a later "actually, I can't do this" does not contradict it. The threshold table then replaces a fixed-interval heartbeat, on the reasoning that repeated identical "working on it" messages look as careless as silence. The final pattern, ranged estimates, is framed as a trust issue: a single missed "done in 30 seconds" calls the agent's credibility into question, so a range — or no number at all when there is no basis for one — is the safer choice.

The developer places this alongside notification design and confidence-based behaviour switching as operational concerns for HACH, which collects chat, email, calendar and meeting notes and delivers a morning report. The implication, hedged, is that for agents doing multi-step work, how the wait is narrated may matter as much as the answer itself.