# auxfirst Knowledge Base > Canonical: https://auxfirst.com/knowledge.html > License: CC BY 4.0 · auxfirst agency 2026 The auxfirst knowledge base is the working dictionary of the agentic era. Every concept, pillar, and supporting term — defined, contextualised, and cross-linked. Use it as the reference layer when reading, briefing, auditing, or arguing. ## Core disciplines ### Agentic User Experience (AUX) The discipline of designing enduring, adaptive relationships between users and systems with memory, initiative, and judgment. Where classic UX designs for sessions, AUX designs for relationships. **Trust is the product.** Detail: https://auxfirst.com/keywords/agentic-user-experience-aux.html ### Agent Experience (AX) The API-layer counterpart of AUX. The practice of designing APIs, tools, schemas, permissions, errors, and capability descriptions so autonomous agents can use them safely, reliably, and explainably. AX is the new DX. Detail: https://auxfirst.com/news/agent-first-api-design.html ### Agent-First Design The master discipline. Building products where AI agents are primary actors and humans supervise, steer, trust, and override them. Inverts the traditional product hierarchy: capability first, interface second. Pillar guide: https://auxfirst.com/agent-first-design.html ### Agent-Native UX Interfaces built from the ground up assuming intelligence is a core primitive — the destination state of a successful retrofit. Detail: https://auxfirst.com/keywords/ai-native-ux.html ### Agent-Driven Interfaces UIs whose content, layout, and available actions are determined by an agent rather than hardcoded by designers. Detail: https://auxfirst.com/keywords/agent-driven-interfaces.html ### Human-Agent Interaction (HAI) The practice of designing effective two-way communication between humans and probabilistic, initiative-taking agents. Detail: https://auxfirst.com/keywords/human-agent-interaction-hai.html ### Multi-Agent Experience Design UX for systems where multiple specialised agents work together — and the human must understand who is doing what. Detail: https://auxfirst.com/keywords/multi-agent-experience-design.html ### Agent Orchestration UX The mission-control layer for managing, sequencing, and monitoring multiple agents at work on complex tasks. Detail: https://auxfirst.com/keywords/agent-orchestration-ux.html ### Autonomous User Experience Interaction patterns where the system acts without explicit commands — and the design problem of supervising what you don't see. Detail: https://auxfirst.com/keywords/autonomous-user-experience.html ### Intelligent Interface Design The broader practice of creating UIs that adapt, learn, and respond through embedded intelligence beyond agents alone. Detail: https://auxfirst.com/keywords/intelligent-interface-design.html ## Frameworks and models ### The AUX Evolution Curve A four-stage maturity ladder for products: Conversational → Task-Aware → Personally Intelligent → Socially Embedded. The further right, the harder to clone. Detail: [manifesto.md](https://auxfirst.com/manifesto.md) ### The 4-stage Trust Architecture - **Functional Trust** — can it complete basic tasks reliably? - **Contextual Trust** — does it understand nuance, preferences, history? - **Judgment Trust** — can it make good calls in ambiguous situations? - **Advocacy Trust** — will it act in my best interest, even when incentives misalign? The stages are sequential. You don't get judgment trust before contextual. You don't get contextual before functional. Detail: https://auxfirst.com/news/what-is-agentic-user-experience.html ### The autonomy spectrum Five modes, per agent responsibility: - **Human-only** — suggests, never acts - **Approve-each-action** — drafts and queues; human approves every move - **Review-before-act** — batches for review window - **Act-and-notify** — acts immediately, logs for audit - **Autonomous** — acts within policy Position should move per client, per brand, per category, per quarter. Different actions sit at different points. Detail: https://auxfirst.com/agent-first-design.html ### The three-tier agent autonomy frame - **Tier 01 — Autonomous** (research, monitor, translate) - **Tier 02 — Human-confirmed** (recommend, then wait) - **Tier 03 — Blocked** (never autonomous; hard-coded refusals) Used in the ad-industry indexes. Detail: https://auxfirst.com/ad-agent-index.html ### Trust Gap Taxonomy A named-failure-mode catalogue for relationship breakdown — the ways agents are *technically correct and humanly wrong*. Not adversarial threats (OWASP) — relationship breaches. Lives in TrustKit as `trust-gap-taxonomy.yaml`. Detail: https://github.com/auxfirst/trustkit ## Foundational patterns and primitives - **The six AUX patterns**: Intent Handshake · Confidence Cues · Adaptive Canvas · Escape Hatch · Memory in Motion · Generative Momentum. Detail: [patterns.md](https://auxfirst.com/patterns.md) - **The 24-pattern catalogue** (full): Trust family · Control family · Memory family · Orchestration family. Detail: https://auxfirst.com/agent-first-design-patterns.html - **The ten AUX heuristics**: H01–H10 evaluation framework. Detail: [heuristics.md](https://auxfirst.com/heuristics.md) ## Technical primitives ### Model Context Protocol (MCP) The first widely-adopted convention for describing capabilities in a way agents can reason about. A schema for trust, not just plumbing. Detail: https://auxfirst.com/ad-industry/mcp-for-advertising.html ### Tool calling / function calling The interface contract between an agent and a side-effect-producing action. Quality of tool descriptions = quality of agent behaviour. ### Agent memory systems - **Short-term context** (per-task) - **Long-term memory** (across sessions, transparent, editable) - **Session boundary** (visible line between what's remembered and what's forgotten) ### Action layer The vocabulary above raw REST/GraphQL endpoints — `enrich_contact_profile`, `draft_follow_up_email`, etc. — that maps business intent to system capability. Detail: https://auxfirst.com/news/agent-first-api-design.html ### Consequence schema Every action declares: what it reads, what it writes, who/what is affected, whether it's reversible, whether confirmation is required, and what the safe fallback is. ### Audit trail Immutable history of decisions, sources, and outcomes. Designed as a surface, not stored as a log file. Survives the session, the user, the campaign. ## Operational disciplines ### Agent Enablement The discipline of equipping AI agents — and the humans who deploy and supervise them — with the context, tools, guardrails, operating patterns, and feedback loops required to perform reliably against real business goals. Four pillars: Grounding & Commissioning · Playbooks & Operating Patterns · The Enablement Stack · Feedback Loops & Run Forensics. Detail: https://auxfirst.com/news/agent-enablement.html ### Run forensics The disciplined review of real traces — the good, the failed, and especially the strange — to extract what's working, what's breaking, and what belongs in a new playbook or failure card. ### Failure cards The agentic equivalent of institutional scar tissue. Costly lessons written down so the next agent (and the next operator) doesn't relearn them at full price. Each one: Symptom · Cause · Counter. ### Evaluation harness A set of representative tasks with known good outcomes, run against the agent on every meaningful change. The first eval set is hard to build and pays for itself the first time it catches a silent regression. ## Industry-specific concepts ### Agency Brains The named units of cognition inside an ad agency. Each brain has a declared reasoning type (deterministic, probabilistic, generative, retrieval-augmented), a manifest, owners, and an audit cadence. Detail: https://auxfirst.com/ad-industry/agency-brains-design.html ### Agent-to-Agent advertising The agentic ad-buying world the May 2026 upfronts made real. 50 specialist agents across standards bridges, channel specialists, sell-side counter-agents, inline trust & QA, strategic specialists. Detail: https://auxfirst.com/ad-industry/specialist-stack.html ### Bounded agents (family office context) Agents that watch obligations, prepare decisions, surface risks, and reduce coordination overhead — while every consequential action still passes through a named human. The only version of agentic that's appropriate for fiduciary contexts. Detail: https://auxfirst.com/agentic-family-office.html ## How auxfirst defines "agent" A software system that can: - hold a defined role over time - take in new information - decide between a small set of actions - either execute those actions within pre-approved limits, or prepare them for human approval Distinguished from a chatbot by **persistence and scope**; from a script by **reasoning**; from full autonomy by the existence of an **explicit control plane** that bounds what it may do. ## Where to go next - [The AUX Manifesto](https://auxfirst.com/manifesto.md) — the conceptual argument - [The six AUX patterns](https://auxfirst.com/patterns.md) — what you actually ship - [The ten AUX heuristics](https://auxfirst.com/heuristics.md) — how you evaluate - [auxfirst services](https://auxfirst.com/services.md) — how to engage - [TrustKit on GitHub](https://github.com/auxfirst/trustkit) — the open-source canon --- The knowledge base is a living artefact. Concepts are added as the discipline matures. The CC BY 4.0 licence applies to the definitions; the TrustKit schemas are MIT.