# The 10 AUX Heuristics > Canonical: https://auxfirst.com/heuristics.html > License: CC BY 4.0 · auxfirst agency 2026 The AUX answer to Nielsen Norman Group's classic usability heuristics — reimagined for agents. Traditional UX heuristics optimise for clarity of interface. AUX heuristics optimise for **quality of relationship**. Use these as an evaluation framework. Run a shipped AI feature through them and you'll find at least three trust gaps you didn't know you had. ## H01 · Visibility of Agent Intent & Action Does the agent make its intent legible *before* or *during* action? Intent playback, plan preview, *"here's what I'm about to do…"* A user should always be able to answer: what is this agent trying to do right now, and what is it about to do next. ## H02 · Progressive Transparency Early: full reasoning, sources, confidence. Later: concise or silent. The trajectory is **transparency → summary → confident → silent**. Match the depth of explanation to the maturity of the relationship. Over-explain in week one. Barely explain in month six. Like a human colleague. ## H03 · User Control Through Steering Do users guide via intent and correction — approve, modify, redirect — not step-by-step control? UX was about *control*. AUX is about *steering*. Users should be able to point the agent at a goal and course-correct, not click through every step. ## H04 · Trust Is Dynamic, Not Static Does agent behaviour evolve as trust develops? New users get cautious, gated behaviour. Experienced users get faster, more autonomous behaviour. Trust is not a switch. It's a curve. The agent's posture should move along the autonomy spectrum as evidence accumulates — and back down when it fails. ## H05 · Clear Boundaries of Autonomy At any moment, does the user know what the agent can do autonomously, what requires confirmation, and what is blocked? **Ambiguity here is where trust collapses.** The autonomy spectrum should be visible, not buried in settings. Users should be able to answer: what can this agent do right now, without me? ## H06 · Graceful Handling of Uncertainty When unsure, does the agent ask, escalate, or return a partial result — not hallucinate? Failure is a first-class design surface, not an exception path. The agent that says "I don't have the grounding for that" is a more trustworthy agent than the one that confidently makes something up. ## H07 · Appropriate Assertiveness Does the agent challenge users when necessary? The spectrum: **compliant → advisory → assertive → protective**. Mature agents push back when something seems wrong. They interpret intent. They flag bad inputs. The agent that just complies is not a collaborator — it's a yes-machine, and that's a future failure mode. ## H08 · Context Efficiency & Awareness Does the agent use context intelligently, not exhaustively? Routing, indexing, and summarisation beat full scans. Context is a cost (tokens, latency, cognitive load) and a performance constraint. An agent that drowns in irrelevant context reasons worse, not better. ## H09 · Multi-Actor & Multi-Agent Clarity Is it always obvious who is acting — the human, this agent, a sub-agent? Who delegated? Who owns the outcome? Multi-agent systems collapse fast without this. Every action should be attributable to a named actor with a named role. ## H10 · Consistency of Behavior, Not Interface The interface can adapt (generative UI). But the agent's **decision logic, trust boundaries, and tone** must stay stable. Consistent agent, adaptive UI — that's the goal. Traditional consistency = same buttons in same places. AUX consistency = same judgment in same situations, regardless of how the surface renders. ## How to use these Treat the ten as a checklist for any agent-facing feature, before and after ship. - **Before ship** — does the design have an explicit answer for each heuristic? If H05 is ambiguous, the surface is not done. - **After ship** — run the trace review (the weekly run-review ritual from the Agent Enablement playbook). Classify every failure against a heuristic. The histogram tells you which one to invest in next. ## Where these live in the open source The heuristics are also published as machine-readable YAML in **TrustKit** — `schemas/aux-heuristics.yaml`. Definitions are CC BY 4.0; the schema is MIT. https://github.com/auxfirst/trustkit ## Background Berkeley's CLTC research on human–AI agent interaction confirms where failures cluster in production: loss of user control and transparency gaps — precisely H03, H05, and H06. These three are usually where the biggest unrealised wins live. --- Related: [The AUX Manifesto](https://auxfirst.com/manifesto.md) · [The six AUX patterns](https://auxfirst.com/patterns.md) · [The complete pattern catalogue](https://auxfirst.com/agent-first-design-patterns.html)