AX (Agent Experience)

AX (agent experience) describes how well a codebase's environment is set up for a coding agent to do good work in it, as distinct from how well it is set up for a human. Matt Pocock's AI Coding Dictionary (aicodingdictionary.com), a glossary of agentic-coding terms open-sourced on GitHub, frames AX as the agent-facing counterpart to DX, developer experience. When the same agent, running the same model and harness, performs well in one repository and badly in another, the dictionary attributes the gap to AX rather than to the model. The instinct to blame the model or rewrite the prompt usually points at the wrong fix.

The dictionary lists three dimensions of good AX. Automated checks, fast and deterministic types, tests, and lints, let an agent self-correct without a human. Architecture matters too: a codebase the agent can navigate without reading everything, with predictable structure and behavior behind small, clearly named interfaces, reduces how much it must hold in mind at once. Free context, keeping AGENTS.md, skills, and tools lean, leaves more of the context window available for the actual task and keeps the agent in what the dictionary calls the smart zone, rather than the degraded state that comes from an overloaded context.

AX and DX overlap, since good checks and clean architecture help both audiences, but they diverge in what each tolerates. Humans absorb tribal knowledge, slow CI, and an informal "ask a colleague." Agents cannot fall back on either, and they gain nothing from IDE tooltips or dashboards built for a person's eyes. A codebase can have good DX and poor AX at the same time. The dictionary treats AX as a distinct property that needs its own investment, not a synonym for DX.