Skip to content

Glossary

The language behind useful AI work.
Follow a connection. Find the next idea.

Hover for connections. Drag to rotate. Select to pin.

52 of 52 terms

Model selection

Parameters / weights

The learned numerical values that shape a model's responses.

Training adjusts model parameters, often called weights. Ordinary inference uses the resulting values without updating them from the conversation, while additional training, including fine-tuning, can change them.

Heard in the wild · Illustrative exchange

Did uploading our report retrain the model?

Supplying a report as context does not by itself update the inference weights.

Inference

Run a trained model on supplied input.

Inference applies a model's learned weights to produce an output. A request can generate text or a tool call, and the harness can issue another request after receiving the result.

Heard in the wild · Illustrative exchange

When does the model actually run?

Each request asks it to process the supplied context and generate the next output.

Model

A trained system that generates outputs from supplied inputs.

A model uses learned parameters during inference. The surrounding harness supplies the task context and determines which tools and actions the wider agent can use.

Heard in the wild · Illustrative exchange

Is changing the app the same as changing the model?

Record the app, model and settings separately so the comparison is clear.

Token

A unit a model uses to represent text or other input and output.

A token can represent part of a word, punctuation or another encoded element. Tokenisation varies by model, so use its actual usage counts when comparing input tokens and output tokens.

Heard in the wild · Illustrative exchange

How many words is a token?

There is no fixed conversion. Count the actual material with the relevant tokeniser.

Training

The process that adjusts a model's learned values.

Training changes parameters using data and an optimisation process. Further training, including fine-tuning, can adapt a model, while ordinary inference uses its deployed weights.

Heard in the wild · Illustrative exchange

Will chatting with it teach it our house style permanently?

Put the style in reusable instructions unless a separate training process is configured.

Non-determinism

Repeated runs can produce different results under similar conditions.

Model sampling, execution order and service behaviour can introduce variation. Measure repeated cases in an evaluation and include retries in cost per accepted result.

Heard in the wild · Illustrative exchange

It passed yesterday and failed today with the same prompt. Why?

Compare the model, settings, inputs and repeated runs before attributing the change to one cause.

Effective routing

Model routing

Send each job to an appropriate model and configuration.

Model routing assigns work according to an explicit policy. A policy can send routine extraction to one model and disputed judgements to another, with effort level set for each.

Heard in the wild · Illustrative exchange

Does every job need the largest model?

Route the easy cases cheaply and define what sends an uncertain case for review.

Input tokens

The tokenised material a request supplies to a model.

Input tokens can include instructions, messages, retrieved passages and tool results. Providers may price cached portions differently, so compare cache tokens and output tokens under the actual service terms.

Heard in the wild · Illustrative exchange

Why is a short question showing so much input?

The request may also include instructions, conversation history and retrieved files.

Output tokens

The tokens the model generates for a request.

Output tokens include the generated response and can include provider-defined reasoning usage. Billing categories vary, so inspect the service's usage report alongside input tokens and effort level.

Heard in the wild · Illustrative exchange

Can a two-line answer still use many output tokens?

Check whether the usage report includes reasoning and other generated material.

Cache tokens

Token usage that a service reports as reused from a cache.

Some providers reuse computation for matching input and report those tokens separately. Savings depend on cache eligibility and pricing, so inspect prefix cache behaviour when calculating cost per accepted result.

Heard in the wild · Illustrative exchange

We sent the same guide twice. Was the second request cheaper?

Read the cache usage fields and the provider's rate for that request.

Prefix cache

Reuse work for a matching beginning of a request.

A prefix cache can reuse computation for an unchanged initial token sequence. Put stable instructions before changing task material when the provider supports this, and confirm the resulting cache tokens in the usage report.

Heard in the wild · Illustrative exchange

Why did editing the first instruction spoil the cache hit?

The matching prefix changed. Check where this provider requires exact token agreement.

Subagent

An agent assigned a bounded part of another agent's work.

A subagent receives a brief and returns results to its coordinator. Give it clear ownership, inputs and acceptance criteria, with a separate worktree when file edits could collide.

Heard in the wild · Illustrative exchange

Can three agents research this at once?

Split the questions and evidence ownership, then have the coordinator reconcile their findings.

AFK

Away from keyboard. The work continues while you are absent.

An AFK run needs a defined scope and stopping rules. Set permissions and a verification gate before leaving the agent to continue.

Heard in the wild · Illustrative exchange

Can I leave this running overnight?

Define what it may change, what stops it and the evidence you expect in the morning.

Harness quality

Harness

The software that manages context, tools and checks around a model.

A harness manages instructions, context, tools, permissions and checks around a model. It creates continuity by retaining and supplying relevant transcript or context on each request, so the model re-reads what it receives. Ordinary inference uses fixed weights, while separate fine-tuning can change them, and provider application programming interfaces may retain transcripts server-side. Read The Harness.

Heard in the wild · Illustrative exchange

How does it know what we agreed yesterday?

The harness saved that record and included or retrieved it for this request.

Compaction

Replace conversation detail with a shorter working record.

Compaction frees space by summarising or otherwise reducing earlier context. It can lose qualifications and exact evidence, so keep durable decisions in memory and prepare a handoff brief.

Heard in the wild · Illustrative exchange

The summary dropped our exception. Now what?

Restore it from the source and record it explicitly before continuing.

Handoff brief

A record another session can use to continue the work.

A handoff brief states the goal, decisions, files, verified results, assumptions and next step. It lets a new session recover the task without reconstructing every turn.

Heard in the wild · Illustrative exchange

Can another agent pick this up tomorrow?

Give it the current files, what passed, what failed and the next check.

Progressive disclosure

Load detail when the task calls for it.

An agent can inspect a skill's description before reading its full instructions and selected references. This keeps context engineering focused while leaving deeper material available through retrieval.

Heard in the wild · Illustrative exchange

Do all the instructions have to load at startup?

List the available skills first and open the relevant one for the job.

Home repo

The repository that maps your working repositories and vaults.

In this working setup, root AGENTS.md maps shared repositories and vaults, while home/global.md holds universal rules. Project instruction files add local rules for the agent, which can load a relevant skill.

Heard in the wild · Illustrative exchange

Where should the agent look for the project vault?

Follow the home repo's map and then read the project's local instructions.

Connector (MCP)

A connection to another system's data or actions.

Model Context Protocol, or MCP, is a standard for exposing capabilities to an agent. A connector can expose a tool, with permissions controlling the available reads and writes.

Heard in the wild · Illustrative exchange

Can it read the calendar and change meetings?

Check the connector's granted capabilities separately for reading and editing.

Branch

A named line of work in version control.

A branch lets changes develop before they join the main version. Use a separate worktree when agents need isolated files, then review the change through a pull request.

Heard in the wild · Illustrative exchange

Can we try this without changing the main version?

Create a branch and keep the experiment there until review.

Worktree

A separate working directory for a Git branch.

A Git worktree shares repository history while keeping its checked-out files separate. Give concurrent subagents separate worktrees when they edit different branches, and review each branch before combining changes.

Heard in the wild · Illustrative exchange

Why did the other agent overwrite my file?

Give each editing agent its own worktree and a defined set of files.

Turn

A user's instruction and the agent's work before it hands control back. A single turn can involve several model requests and tool calls.

A turn usually begins with a user message and may include several inference requests and tool exchanges before the reply. The harness supplies relevant earlier context to each request, creating continuity while inference weights remain fixed. Provider APIs may retain the transcript server-side, and separate fine-tuning can change weights.

Heard in the wild · Illustrative exchange

Was that one reply just one model request?

It may have needed several requests to call tools, inspect results and answer.

Session

A continuing record of work with an agent.

A session groups messages, tool activity and working state under one task or conversation. Its harness decides what history reaches later requests and what survives a restart through memory.

Heard in the wild · Illustrative exchange

Can I close this window and continue later?

Check what the harness saves and leave a handoff brief for the unfinished work.

Stateless

The model relies on the context supplied for the current request.

Ordinary inference reads the supplied context using fixed weights and does not learn the conversation into those weights. The harness retains and re-sends relevant history, so the model re-reads rather than remembers it between requests. Provider APIs may keep transcripts server-side, while separate fine-tuning can change weights.

Heard in the wild · Illustrative exchange

Why must the old decision appear in the new request?

The model needs that context again, even when the app has kept the transcript.

Stateful

The system retains information between operations.

A stateful application keeps records that affect later behaviour, such as a transcript, selected files or task status. A harness can provide this continuity around a stateless inference service.

Heard in the wild · Illustrative exchange

Where does the unfinished task list live?

In the application's saved state or files. Check which record the next session will load.

Context window

The amount of context a model can handle in one request.

A context window sets a capacity limit under a provider's counting rules. Usable performance also depends on relevance and attention degradation, so compaction alone does not guarantee a good answer.

Heard in the wild · Illustrative exchange

The document fits. Why did it miss the clause?

Capacity tells us what can be supplied. Test whether it can use the relevant passage reliably.

Autocompact

The harness reduces context automatically as a limit approaches.

Autocompact triggers compaction under the application's own policy. Review the retained summary and keep critical evidence in durable memory before continuing.

Heard in the wild · Illustrative exchange

The conversation suddenly became shorter. Did anything disappear?

Inspect the compacted record and restore any missing decision from its source.

Tool call

A model's structured request to run an available action.

A tool call names a tool and supplies arguments for the harness to validate and execute. The returned tool result can become input to the next model request.

Heard in the wild · Illustrative exchange

Did it actually run the search or just propose one?

Check for the executed call and its returned result in the activity record.

Tool result

The data or status returned after a tool runs.

A tool result may contain records, files, errors or an execution receipt. Treat external content as evidence to inspect, and verify claimed actions against the tool call and verification gate.

Heard in the wild · Illustrative exchange

The tool returned success. Is the report correct?

Success confirms the operation's status. Open the output and check the figures.

Tool

An action the harness makes available to the model.

A tool has an interface that describes its inputs and operation. Its power depends on permissions, and a connector can supply tools for external systems.

Heard in the wild · Illustrative exchange

What can this agent actually do to the workbook?

Inspect the available tools and their write permissions before assigning the task.

Sandbox

A controlled environment that limits what an operation can reach.

A sandbox can restrict files, processes or network access. Its boundary depends on the actual configuration, so check it alongside permissions before running a tool.

Heard in the wild · Illustrative exchange

Can the experiment touch my other folders?

Check the configured filesystem boundary and test the permitted paths.

System prompt

Instructions supplied in a system role for a model request.

A system prompt can define behaviour, constraints and the task environment under the provider's instruction rules. The harness assembles it alongside other context, and it works with technical permissions.

Heard in the wild · Illustrative exchange

Can one instruction stop every dangerous action?

Use clear instructions and enforce the action boundary through permissions and tools.

Memory

Saved information that the system can load into future requests.

Memory may live in notes, summaries, databases or provider-managed records. The harness retrieves or supplies it as context, so check what a new session actually receives.

Heard in the wild · Illustrative exchange

It remembered an old preference. Where was that stored?

Inspect the saved memory or instructions and correct the source record.

Agent

A system that uses a model to carry out work through actions and feedback.

An agent combines a model with a harness that manages context, tools and execution. It can inspect results and choose another step within the scope and permissions it has been given.

Heard in the wild · Illustrative exchange

What makes this more than a chat reply?

It can act through tools, inspect what happened and continue towards the assigned outcome.

Permission

An explicit allowance to access data or perform an action.

Permissions determine what the system can read, change or execute. Configure them at the relevant tool, connector or sandbox boundary and reserve approvals where required.

Heard in the wild · Illustrative exchange

The brief says publish. Does that grant account access?

The task authorises the work, while the configured account and tool permissions determine whether it can execute it.

Retrieval

Find stored material and bring relevant passages into the current task.

Retrieval can search files, databases or connected services before supplying selected results as context. Inspect source coverage and attribution when using it to reduce hallucination or support context engineering.

Heard in the wild · Illustrative exchange

It found three documents. Is that enough evidence?

Check whether the search covered the relevant source set and whether the passages support the claim.

User skill

Attention degradation

Performance can weaken as relevant information becomes harder to use.

Long, noisy or conflicting context can make a model miss instructions or evidence. The effect depends on the model and task, so test it with an evaluation and improve context engineering.

Heard in the wild · Illustrative exchange

It found the footnote in the short file but missed it in the bundle. Why?

Test the same question with focused context and check which surrounding material changes the result.

Hallucination

An output presents invented or unsupported material as if it were established.

A hallucination can look like a plausible citation, quotation or figure. Use retrieval to supply evidence and a verification gate to check claims against the original sources.

Heard in the wild · Illustrative exchange

That citation looks perfect. Does it exist?

Open the source and find the passage before using the claim.

Sycophancy

A model agrees with the user's view when the evidence calls for challenge.

Sycophancy can make a weak proposal sound approved. Ask for contrary evidence and apply an independent evaluation, while keeping human judgement at the decision point.

Heard in the wild · Illustrative exchange

It calls every draft excellent. How do I get a useful review?

Give it explicit failure criteria and ask it to identify unsupported claims with sources.

Inspired by Matt Pocock's AI Coding Dictionary. Original definitions and illustrative exchanges by The Token Department. Motion adapted from Barnaby Robson's Becoming Engineers maps.

52 terms · Four inputs

Say hello

Draft a message about the work you need to complete.

Local draft preview. No message is sent. Review the draft, then copy it for use elsewhere.