AX Framework

Agentic Experience 2026

Principles & Framework for Autonomous, Intent-Based Interfaces.

We’re shipping agentic systems — but still designing them like chatbots. This framework proposes a native design language for interfaces where machines act on your behalf.

Download the Framework (PDF)

What’s Inside

Core Principles

Five design patterns for agentic interfaces that go beyond prompt-and-respond. Goal-First Onboarding starts from intent, not instructions. Shared Autonomy introduces the Variable Leash — a sliding scale of human control. Parallel Agency addresses what happens when multiple agents operate simultaneously on your behalf. Outcome Projection uses Ghost States to show you where an agent is heading before it arrives. Progressive Auditability replaces the false binary of black box versus full transparency.

Taxonomy of Agency

A four-class responsibility ladder from micro-tasking to legal persona. Each class has different requirements for authorization, oversight, and accountability — and most teams are designing Class 3 interfaces with Class 1 assumptions. The taxonomy gives product teams a shared vocabulary for scoping what their agents should and shouldn’t do.

Technical & Ethical Standards

Authorization contracts that make agent permissions explicit and auditable. Memory sovereignty principles that put data ownership in the user’s hands. Kill-switch design for graceful degradation, not catastrophic failure. Inter-agent coordination protocols and resource transparency standards that prevent the “agent swarm” problem before it starts.

Social Calibration

The Instrumental–Relational persona scale. Not every agent should feel like a friend, and not every agent should feel like a tool. This section maps when warmth helps and when it deceives — and gives designers a framework for calibrating the social register of their agent’s interactions.

This is a working document, not a finished product. The framework is most useful when it collides with the real problems teams are solving right now.

Open Questions

The hard problems in agentic design don’t have clean answers yet. These are the questions I keep coming back to — and the ones I’d most like to argue about.

When Agent A delegates to Agent B delegates to Agent C, who’s accountable?

When an agent acts “as you,” what identity does it carry?

What’s the right default for memory sovereignty?

Get Involved

This framework is most useful when it’s battle-tested.

If you’re building agentic systems and wrestling with these problems, I’d value your perspective.

Or email directly: james@touchgrass.consulting

Not ready? Subscribe to my writing or follow on LinkedIn