{"slug": "ai-coding-dictionary", "title": "AI Coding Dictionary", "summary": "Matt Pocock published a dictionary of AI coding terminology with 69 entries, hosted on GitHub at mattpocock/dictionary-of-ai-coding. The glossary organizes terms into sections including \"The Model\" (16 terms), \"Sessions, Context Windows & Turns\" (8 terms), and \"Tools & Environment\" (10 terms), defining concepts such as tokens, context windows, harnesses, and agents in plain English.", "body_md": "AI Hero · Dictionary\n\n# The vocabulary of AI coding,\n\n in plain English.\n\nSkimmable definitions for the terms that make agentic coding click. Search 69 entries below, or jump into a section.\n\n[4,569mattpocock/dictionary-of-ai-coding](https://github.com/mattpocock/dictionary-of-ai-coding)\n\n## The Model\n\n16 terms\n\n### AI\n\nA moving label, not a technology. Points at whatever computers can newly, impressively do — right now, large language models.\n\n### Model\n\nThe parameters. Stateless — does next-token prediction and nothing else. Cannot do anything agentic on its own.\n\n### Parameters\n\nThe numbers inside a model — often billions — tuned during training. Everything the model knows lives in them. Also called weights.\n\n### Training\n\nThe process that sets a model's parameters by exposing it to vast amounts of text and adjusting to improve next-token prediction.\n\n### Inference\n\nRunning a trained model to generate output — what happens on every model provider request. Parameters stay fixed.\n\n### Effort\n\nA dial for how much reasoning the model does before it answers. More effort spends more output tokens for a better shot at hard problems.\n\n### Token\n\nThe atomic unit a model reads and writes. Roughly word-sized but not exactly. Context window size, cost, and latency all count tokens.\n\n### Next-token prediction\n\nWhat the model actually does. Samples one next token from the context, appends it, and runs again. Its only mode of operation.\n\n### Non-determinism\n\nThe same input can produce different output. A property of how models generate text and how providers serve requests.\n\n### Model provider\n\nWhatever serves a model for inference. Usually remote (Anthropic, OpenAI, Google), but can also be local (Ollama, llama.cpp).\n\n### Harness\n\nEverything around the model that turns it into an agent: tools, system prompt, context-window management, permissions, hooks.\n\n### Model provider request\n\nOne round-trip from the harness to the model provider. The harness sends context; the provider returns one response.\n\n### Input tokens\n\nTokens the harness sends on each model provider request. Billed at a lower rate than output tokens.\n\n### Output tokens\n\nTokens the model generates back. Billed at a higher rate than input tokens, since they cost more compute to produce.\n\n### Prefix cache\n\nThe provider-side store that lets consecutive requests skip re-processing a shared prefix, billing those tokens at a lower rate.\n\n### Cache tokens\n\nInput tokens the provider has cached from a previous request via its prefix cache, billed at a much lower rate.\n\n## Sessions, Context Windows & Turns\n\n8 terms\n\n### Stateless\n\nCarries no information forward. The model is stateless across requests; an agent is stateless across sessions by default.\n\n### Context\n\nThe relevant information the agent has access to right now — what the agent knows that's pertinent to the task.\n\n### Context window\n\nEverything the model sees on each model provider request. Finite, model-specific, the only surface through which the model perceives.\n\n### Stateful\n\nCarries information forward. Sessions are stateful across turns; agents can be made stateful across sessions via a memory system.\n\n### Agent\n\nA model harnessed with tools, a system prompt, and a context window, that takes turns with a user. The model in motion.\n\n### System prompt\n\nThe instructions the harness prepends to every model provider request — the agent's standing brief. Usually stable across a session.\n\n### Session\n\nOne bounded run of interaction with an agent. Starts empty, accumulates, ends when cleared, closed, or compacted into a fresh session.\n\n### Turn\n\nOne user message plus everything the agent does in response, up until it yields back to the user. Contains one or more provider requests.\n\n## Tools & Environment\n\n10 terms\n\n### Environment\n\nThe world the agent acts on — anything outside the harness that the agent perceives via tool results and changes via tool calls.\n\n### Filesystem\n\nA tree of files and directories the agent reads from, writes to, and executes within — the default environment for a coding agent.\n\n### Tool\n\nA function the harness exposes for the agent to call — Read, Write, Bash, Search. How an agent perceives and acts on the environment.\n\n### Tool call\n\nThe model's output naming a tool and its arguments — just structured text. The harness has to read it and execute.\n\n### Tool result\n\nWhat the harness sends back after executing a tool call — file contents, output, or error. The agent's only view of the environment.\n\n### MCP\n\nA protocol for plugging external tool servers into a harness — how an agent gets tools beyond what the harness ships with.\n\n### Permission request\n\nWhat the harness shows the user before executing a tool call that isn't pre-approved. The mechanism for putting a human in the loop.\n\n### Permission mode\n\nThe permission-gating slice of an agent mode — which tool calls trigger a permission request and which run automatically.\n\n### Agent mode\n\nA preset bundling a permission mode with behavioral instructions injected into the system prompt. Can flip mid-session.\n\n### Sandbox\n\nAn isolated environment the agent runs inside — container, VM, or restricted shell. Limits the blast radius of agent actions.\n\n## Failure Modes\n\n9 terms\n\n### Sycophancy\n\nConfidently agreeable model output. Caused by training that shaped the model to favor answers humans liked — including agreement.\n\n### Hallucination\n\nConfidently-wrong model output. Two flavors: factuality (invented facts) and faithfulness (drift from loaded context).\n\n### Parametric knowledge\n\nWhat the model knows from training, stored in its parameters. Frozen at training time. Counterpart to contextual knowledge.\n\n### Knowledge cutoff\n\nThe date past which a model has no parametric knowledge. Post-cutoff libraries and APIs are fabrication traps unless docs are loaded.\n\n### Contextual knowledge\n\nFacts the agent can read directly from the context right now. Counterpart to parametric knowledge.\n\n### Attention relationship\n\nThe pairing between two tokens — meaningful pairs influence each other more than unrelated ones. A context of N tokens has ~N² of these.\n\n### Attention budget\n\nEach token has a finite amount of influence to distribute across the rest of the context. Per-token, doesn't grow when context does.\n\n### Attention degradation\n\nAs a session grows, each token's attention budget spreads across more competitors; signal on meaningful relationships shrinks.\n\n### Smart zone\n\nEarly in a session the agent is sharp and focused. As the session grows it drifts into a dumb zone: sloppier, forgetful, more mistakes.\n\n## Handoffs\n\n9 terms\n\n### Clearing\n\nEnding the current session and starting a fresh one. The next message begins with an empty session and an empty context window.\n\n### Handoff\n\nTransferring agent context from one session to another, with no return path. Carry mechanism varies — artifact, compaction, others.\n\n### Primary source\n\nThe thing itself — code, transcripts, raw data. Complete and authoritative, but expensive to load into context.\n\n### Secondary source\n\nAn account of a primary source, one step removed — summaries, docs, compaction summaries. Cheap to load, lossy by construction.\n\n### Handoff artifact\n\nA document used as the carry mechanism for a handoff — written by one session to be read by another.\n\n### Spec\n\nA handoff artifact describing a multi-session piece of work — what's being built, not how each session does its share. Made of tickets.\n\n### Ticket\n\nA handoff artifact scoping one session of work. Stands alone or hangs off a spec. Can block or be blocked by sibling tickets.\n\n### Compaction\n\nA handoff done in-memory: the previous session's history is summarised and seeds a fresh session. Lossy — detail traded for headroom.\n\n### Autocompact\n\nCompaction triggered automatically by the harness when the context window approaches full.\n\n## Memory and Steering\n\n6 terms\n\n### Memory system\n\nA system that attempts to make an agent stateful across sessions by persisting to the environment and reloading at session start.\n\n### AGENTS.md\n\nA file in the environment that the harness loads into the context window at session start — the project's standing brief to the agent.\n\n### Progressive disclosure\n\nLoading only the context an agent needs right now, with context pointers to the rest. Borrowed from UI design.\n\n### Context pointer\n\nA mention in one document that points to another, so the agent can pull it into context only when the task calls for it.\n\n### Skill\n\nA teachable capability bundled as a unit — kept out of the context window until a context pointer pulls it in for the task at hand.\n\n### Subagent\n\nAn agent spawned by another agent via a tool call. Runs in its own session, reports a single tool result. Cannot spawn further subagents.\n\n## Patterns of Work\n\n11 terms\n\n### Human-in-the-loop\n\nA working pattern where one or more humans pair with the agent during a session — reviewing, redirecting, or collaborating in real time.\n\n### AFK\n\nA working pattern where the user kicks off a session and leaves the agent to run unattended (away from keyboard).\n\n### Automated check\n\nA deterministic verification that runs in the environment — tests, type checks, lints, build, pre-commit hooks. Pass/fail, no judgement.\n\n### Automated review\n\nAn agent reviewing another agent's work, often with a different model or system prompt. Non-deterministic: it forms a judgement.\n\n### Human review\n\nThe user reading the code the agent produced and forming a judgement on it. Reading the diff counts; reading the summary doesn't.\n\n### Vibe coding\n\nA working pattern where the user accepts the agent's code without human review. The diff is treated as opaque.\n\n### Design concept\n\nThe shared understanding of what's being built, held in common between user and agent but separate from any asset.\n\n### Grilling\n\nA technique for developing a design concept: the agent interviews the user Socratically, one decision at a time.\n\n### Prototyping\n\nHaving the agent build a quick, rough version when conversation is too low-fidelity and you need a real artifact to talk about.\n\n### DX\n\nDeveloper experience: how easy a codebase and its toolchain make it for humans to do good work — docs, feedback speed, errors.\n\n### AX\n\nAgent experience: how well the environment is set up for an agent to do good work — checks, architecture, and free context.", "url": "https://wpnews.pro/news/ai-coding-dictionary", "canonical_source": "https://www.aihero.dev/ai-coding-dictionary", "published_at": "2026-09-13 23:56:24+00:00", "updated_at": "2026-09-14 00:27:33.753238+00:00", "lang": "en", "topics": ["ai-tools", "developer-tools", "large-language-models", "ai-agents"], "entities": ["Matt Pocock", "GitHub", "Anthropic", "OpenAI", "Google", "Ollama", "llama.cpp"], "alternates": {"html": "https://wpnews.pro/news/ai-coding-dictionary", "markdown": "https://wpnews.pro/news/ai-coding-dictionary.md", "text": "https://wpnews.pro/news/ai-coding-dictionary.txt", "jsonld": "https://wpnews.pro/news/ai-coding-dictionary.jsonld"}}