We propose Artificial eXperience Intelligence (AXI) as a name and a structure for an integrating discipline whose unit of optimization is the quality of a human being’s lived experience of an AI system, rather than the system’s raw intelligence, autonomy, or throughput. AXI is positioned as a bridge between artificial intelligence and the humans who use it — a proposed discipline for helping the capability delivered by AI systems translate into a lived experience that is present, embodied, comfortable, and trustworthy for the human on the other side. Across a decade in which generative models reached approximately 900 million weekly active users on ChatGPT alone [1] and the two leading frontier AI companies together exceeded $55 billion in annualized revenue by early 2026 [2], operational experience data tells a different story: only 46% of global respondents are willing to trust AI despite 66% using it regularly [3]; 95% of enterprise generative-AI pilots produce zero measurable P&L impact [4]; Gartner forecasts over 40% of agentic-AI projects will be canceled by 2027 [5]; and a majority of consumers report preferring alternatives to poorly-integrated AI interactions [6]. Capability has advanced; experience has not kept pace. AXI is proposed as a framework for closing that gap, sitting alongside and building on existing disciplines — Artificial Intelligence, Human-Computer Interaction, Explainable AI, Multimodal AI, Embodied AI, Conversational AI, Affective Computing, and Human-Centered AI — while adding a single integrating objective: the quality of the human’s felt experience. We outline a proposed five-layer reference architecture (the AXI Stack), nine design priorities (the AXI Principles), and seven indicative metrics (the AXI Evaluation Framework). We illustrate the proposal through Let Me Teach — the author’s reference implementation — and invite the research community to critique, extend, and refine the framework.


Executive Summary

1. The experience layer is an emerging frontier. The past decade optimized for capability — parameter counts, benchmark scores, agent autonomy. The decade ahead may benefit from parallel attention to the user’s lived experience of AI systems. The gap is documented: 51% of US adults are more concerned than excited about AI [7], trust in AI companies to protect personal data declined measurably between 2023 and 2024 [8], and AI-related incidents rose 56.4% in a single year [8].

2. AXI is proposed as a bridge between AI and humans. Artificial eXperience Intelligence is offered as a discipline that architects an AI system end-to-end so that end users perceive and interact with it as present, embodied, comfortable, and trustworthy participants in their real-world context. AXI is not a replacement for AI, AGI, HCI, or XAI. It is proposed as a bridge that helps their capability reach the human, and as an integrating layer that sits alongside them.

3. Building on prior work. AXI builds directly on Amershi et al.’s Guidelines for Human-AI Interaction [27], Shneiderman’s Human-Centered AI [28], and Xu’s UX 3.0 paradigm for human-centered AI systems [61]. Our contribution is an attempt to name and structure an integrating layer — not to replace or dismiss these foundational efforts.

4. Five layers. The AXI Stack comprises (1) a Perception Layer, (2) a Cognition Layer, (3) an Action Layer, (4) an Embodiment Layer, and (5) an Experience Layer. The first four are addressed by existing disciplines. The fifth is the integration point we propose to formalize.

5. Nine design priorities. Presence over power. Pacing over speed. Transparency over opacity. Continuity over novelty. Comfort over capability. Consent over convenience. Repair over perfection. Proximity over omniscience. Humility over overreach. These are design priorities under resource constraints, not absolute hierarchies.

6. Seven indicative metrics. AXI proposes operationally-measurable indicators for the experience layer, offered as starting points for community calibration rather than empirically-validated thresholds.

7. Reference implementation. Sripto Corporation’s Let Me Teach (launched 2026) is offered as a candidate reference implementation — an interactive explainer that teaches any topic in real time. The case-study observations are preliminary, non-independent, and illustrative, not evaluative.

8. Open research. This paper is published under CC BY 4.0. The framework is offered to the community for adoption, critique, and extension.