What related resource would be useful for someone trying to understand broader AI design changes after reading this article?

A useful next read is “AI-Driven UX Design: Future Roles and Skills for Success,” which says AI is reshaping how designers create, collaborate, and innovate. It fits the primary article’s theme because both focus on how design practice is changing in an AI-driven world.

All insights

Briefing · AI / Orchestration / Operations

Eight Heuristics for Generative and Agentic AI Products

Written by: Pavel Bukengolts
Published: April 20, 2026
Reading time: 2 min read

Topics

Summary

The article says generative and agentic AI fail in ways classic UX frameworks do not fully catch, such as believable hallucinations, hidden irreversible actions, and accessibility breakdowns.
It argues that Nielsen, Microsoft HAX, and Google PAIR still matter, but teams need an updated framework built specifically for current AI systems.
UX Design Lab proposes eight heuristics with clear definitions, failure modes, runnable tests, and pass/fail thresholds to make review more evaluable.
The framework adds practical mechanics like scaling confirmation friction to action risk, measuring the real cost of mid-task AI corrections, and testing accessibility under live streaming and async updates.
It also includes a governance layer that maps compliance requirements to heuristics, while openly noting weaker spots such as qualitative tone scoring and interpretive latency thresholds.
The document is a v0.25 working draft, and the authors invite real-world feedback to strengthen the framework further.

FAQ · 7

What is the main takeaway of “Eight Heuristics for Generative and Agentic AI Products”?

The article argues that generative and agentic systems fail in ways classic UX frameworks were not built to catch. It says Nielsen’s heuristics, Microsoft’s HAX guidance, and Google’s PAIR guidebook still matter, but are not enough on their own, so UX Design Lab extended them into something evaluable for current systems.

What does the article say about generative and agentic systems that makes them different from classic UX problems?

It highlights failures like a hallucinated citation that looks identical to a sourced answer, or an agent taking an irreversible action the user never saw coming. It also notes that a screen reader can lose its place every time the model finishes a sentence, showing why standard UX frameworks miss these issues.

What concrete structure does the article say the eight heuristics include?

Each heuristic comes with a definition, a failure mode, a test you can actually run, and a pass/fail threshold. That makes the framework explicitly evaluable rather than purely descriptive.

How does the article connect to the idea of evaluation literacy in AI-era design work?

Taken together, the primary article and “From UX Artifact Maker to AI System Orchestrator” both point toward a shift in what designers need to do: not just make artifacts, but evaluate AI systems. The secondary article frames this as “evaluation literacy” becoming a valuable skill as work becomes easier to generate but harder to trust.

Grounded in related UXDL articles:

From UX Artifact Maker to AI System Orchestrator

What practical lesson should readers take away from the primary article and HAX guidance?

The primary article says classic guidance still matters, but needs to be extended for current AI systems; the HAX article similarly focuses on the promises and pitfalls of AI in UX design. Together, they suggest designers should look for AI-specific breakdowns instead of assuming traditional UX checks will catch them all.

Grounded in related UXDL articles:

Human-AI Interaction (HAX): Enhancing UX Design for Better AI Experiences

What pitfall or misconception does the article warn against?

It warns against assuming classic UX frameworks are sufficient for generative and agentic systems. The article specifically says those frameworks still matter, but “they are just not enough on their own.”

Generative and agentic systems break in ways that classic UX frameworks were not built to catch. A hallucinated citation can look identical to a sourced answer. An agent can take an irreversible action that the user never saw coming. A screen reader can lose its place every time the model finishes a sentence. Nielsen's heuristics, Microsoft's HAX guidance, and Google's PAIR guidebook still matter. They are just not enough on their own.

Timeline diagram showing major UX, accessibility, and human-AI frameworks that informed UX Design Lab’s Eight Heuristics (2026). It lists Jakob Nielsen’s 10 Usability Heuristics (1994), Jill Gerhardt-Powals’ Cognitive Engineering Principles (1996), Eric Horvitz’s Principles of Mixed-Initiative User Interfaces (1999), WCAG 1.0 (1999), DARPA Explainable AI (2016), Google’s People + AI Guidebook (2019), Microsoft’s Guidelines for Human-AI Interaction (HAX) (2019), OECD AI Principles (2019), and NIST AI RMF 1.0 (2023), all converging into UX Design Lab Eight Heuristics, 2026. — Fig. 01Timeline of UX, accessibility, and human-AI frameworks from 1994 to 2023 converging into UX Design Lab’s Eight Heuristics, 2026.

At UX Design Lab, we took what those frameworks established and extended them into something evaluable for the systems being built right now. Eight heuristics, each with a definition, a failure mode, a test you can actually run, and a pass/fail threshold. The mechanics that make them different from prior guidance include:

Scaling preview friction to the blast radius of an action, so high-impact agent behavior gets a full abort path and low-impact actions run quietly — without training users to click through without reading
Measuring the actual time and quality cost of redirecting an AI mid-task, rather than treating correction as equivalent to starting over
Testing dynamic accessibility under live streaming and async updates, not just against a static DOM scan that will miss every failure that matters in a generative interface
A governance companion layer that crosswalks compliance requirements back to specific heuristics, so legal risk doesn't get siloed away from the UX review

There is also a known soft spot we are actively working on: the tone evaluation in Heuristic 7 is intentionally qualitative, and the latency thresholds in Heuristic 8 are tiered to human perception but still interpretive. We named them in the document rather than hiding them. That is where the most useful pushback will come from.

This is v0.25, a working draft. The PDF gives the short version. The full document is linked below, view-only, no form.

If you use it on a real product and find a weak spot, a missing failure mode, or a place where it breaks under pressure, reach out. We want to build a stronger version with people who are actually doing the work.

Read the full framework in the Google Doc: Eight Heuristics for Generative and Agentic AI Products
Download the short PDF version: PDF