Observed Signal · May 18, 2026 · Analysis · Source: UX Collective · Impact: 2/5 · Sentiment: Negative
AI Products Fail the Doherty Threshold
An analysis by Adi Leviim argues that modern AI chat products and agentic systems routinely violate long-established HCI response-time conventions — notably the 1982 Doherty Threshold (~400 ms) — causing user attention to leak and prompting coping rituals (tab checks, reloads, ‘are you there?’ prompts, screen recording). The author presents measured latency bands for chat and agent operations (from sub-second token streaming to multi-hour async tasks), critiques current feedback affordances (ellipsis, pulsing dots, sparse agent status), and outlines UX conventions that AI products should adopt: continuous progress indicators, updating ETAs, OS-level completion notifications, and persistent readable logs. Leviim frames the waiting problem as a design failure rather than a technical limitation and ties the solution to decades-old OS and long-running-operation UX patterns.
Highlights a widespread UX/design gap in AI chat and agent products that affects user attention, trust, and adoption; implications for product teams and platform UX conventions but not an immediate platform policy or technical-release event.
Track LinkedIn Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- The Doherty Threshold (Walter J. Doherty, 1982) is roughly 400 milliseconds; below it users remain productive.
- Author-provided latency bands: chat token first byte 0.5–3s; full chat response 10–30s; reasoning model response 30–90s; short agent task 2–5 minutes; long agent task 20 minutes–1 hour; background async task 1–3 hours.
- The article states major AI chat products launched in 2023–2024 missed the Doherty Threshold; 2025–2026 agentic systems produce response times measured in minutes to hours.
- Recommended AI waiting UX conventions: continuous progress indicators (not just a pulsing dot), updating time-remaining estimates, OS-routed completion notifications, and persistent readable logs accessible during and after runs.
Connected Companies & Entities
2 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Designing AI Interfaces: Split Acknowledgment and Answer Budgets
This article argues that the root cause of perceived AI sluggishness is not model inference latency but the lack of a separate acknowledgment budget for the user interface. Based on Doherty's threshold of 400ms, it advocates for measuring and designing two distinct clocks: one for interface feedback (acknowledgment) and one for the final answer. It suggests practical design patterns like echoing user input, showing real system state, and auditing agent steps to prevent users from abandoning oversight, thereby preserving the human-in-the-loop needed for AI reliability.
Chat Box Shipped Fast — Not the Right UI
Adi Leviim argues that the ubiquitous AI chat box became the default interface because it was the fastest to ship with large language models, not because it is the best UX. The essay reviews decades of HCI patterns (direct manipulation, structured forms, progressive disclosure) lost to the chat-only surface and documents 2024 product retrofits from OpenAI, Anthropic and Google that reintroduced GUI surfaces (GPT Store, GPT-4o Voice, Artifacts, Projects, Canvas, Computer Use, Deep Research). Leviim distinguishes intent-based interaction from chat-based implementation and predicts the next phase of AI UX will be many small, scoped, structured surfaces (inline rephrase, highlight-to-rewrite, one-click variants) rather than general-purpose chat windows.
Prompt-Only Empty States Hurt AI Product Retention
Adi Leviim argues that modern AI products have replaced decades of empty-state UX design with a single centered prompt box and suggestion chips, which increases first-session drop-off. Drawing on A/B tests and anonymized analytics from a Chrome extension he co‑built, Leviim reports roughly 70% of new installs never return for a second session; versions that showed worked examples or pre-populated artifacts materially outperformed blank-prompt variants. The essay traces this failure to abandoned HCI principles (signifiers, recognition over recall, scaffolding) and recommends four properties for a deliberately designed AI empty state: show a worked example, offer a starting verb, expose model limits, and enable action before the first prompt. The author discloses commercial interest in extensions that fill AI UX gaps.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.