Author(s): Ankit Agrawal

Originally published on Towards AI.

Latency masking for agent UIs: what six decades of perception research — and a few years of shipping chatbots — taught me about designing the wait.

So here’s a scene I’ve watched play out more times than I’d like to admit.

The author argues that in agent and chatbot interfaces, latency is first and foremost a perception problem: empty, uncertain, unexplained waits (especially before work visibly starts) feel longer and less trustworthy, and classic research on waiting lines and response-time thresholds translates directly into UI rules for “latency masking.” While streaming helps by turning idle time into reading time, streaming alone fails for agents whose work includes multi-phase tool calls and “silent gaps” before any token appears, so the product must also frame what’s happening—e.g., show honest step-level status or tool-call transparency rather than bare spinners. The piece highlights counterintuitive findings that faster can sometimes feel worse because delays can be read as deliberation, so latency tolerance depends on the task and what the wait is meant to signify. It recommends designing the wait with the same intent as the answer: avoid theatrical or deceptive progress signals, provide clear progress and control (including interruption/cancellation behavior), and ensure the user can always regain control rather than being trapped by silence; ultimately, the wait becomes part of the product experience and supports user trust even as agent workloads grow.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI