Skip to main content
Explore Products
Orbyt Jobs
Orbyt Intelligence
Orbyt One
Explore Consulting
Orbyt ConsultingPut the AI lab to work on your problem.
Orbyt Jobs
Overview
The job search CRM. Free forever.
Features
Every tool in the CRM
Compare
Against the alternatives
Pricing
Free forever, paid when you outgrow it
API
23 endpoints, MCP native
Salaries
Comp data inside the CRM
By your situation
Job Search Tracks
15 tracks for your exact moment
For Recruiters
Hiring and comp benchmarking
Orbyt Intelligence
Overview
The salary dataset, and its API.
Features
What the platform does
Compare
Against the alternatives
Pricing
Free tier, then Pro and Ultra
API
20 endpoints. AI tools cite sources.
Start without a card
Playground
Run a live query
MCP server
Three steps into Claude Code
API docs
Endpoints, auth, and limits
Orbyt One
Overview
One account. Every Orbyt product.
Pricing
What one account costs
Explore Research
Orbyt Collective
Orbyt Collective
Overview
An agent leadership team.
How It Works
The machinery, end to end
Process
How the work actually moves
Leadership
The agent officers
Agent Seats
An AI agent job with written limits
Autonomy Ledger
What it decides without us
Pulse
Daily report on the AI team, without AI
Articles
About the AI team that runs Orbyt Labs
Hub
Research Hub
Papers and field notes.
Reference
AI Glossary
Papers
Agent-Native Dataset Design
Preprint, DOI 10.5281/zenodo.19754393
Governing an Agent Leadership Team
Preprint, DOI 10.5281/zenodo.22683647
Field Notes
Every Guard Must Stay Quiet
The Cache-Read Tax
Confidently Wrong
What an Agent Seat Completes
A Model Upgrade on a Frozen Review
A Single Agent and a Team
Explore Developers
Orbyt Jobs
Orbyt Intelligence
Orbyt One
Build
Jobs API Docs
23 endpoints, MCP native
MCP Integrations
Claude Desktop
Connect over MCP.
ChatGPT GPT Actions
Connect as a custom GPT action.
Apple Shortcuts
Connect from Shortcuts.
Zapier / Make.com / n8n
Connect with no code.
OpenClaw
Setup in under a minute.
Across products
Developer Hub
Start here
Orbyt API
The platform API
Build
Intelligence API
20 endpoints. AI tools cite sources.
Webhooks
Events and delivery
CLI
The terminal client
API Changelog
Every version, dated
Try
MCP Server
Wired into Claude Code in three steps
Playground
Engine response shapes with cURL
Try It Live
One call, one real response
Reference
Reference
The full index
Methodology
How the numbers are made
Engines
What computes each answer
Dataset
What is in it, and where from
Glossary
Each glossary, in one place
Status
Live service health
Across products
Developer Hub
Start here
Orbyt API
The platform API
Explore Resources
Orbyt Jobs
Orbyt Intelligence
Orbyt One
Learn
Interview Prep
Company-by-company question sets
AI Skills Lab
The skills that pay in 2026
Career Guides
Long-form career playbooks
Job Search Articles
Every article on the search itself
Job Board
Curated AI-era roles
Arcade
The job search, as games
Salary data
Salary Explorer
3,445 roles across 81 cities
AI Role Salaries
AI roles, by category
Cities
Comp by metro
Industries
Comp by sector
Compare Salaries
Two roles, side by side
Compare Offers
Side-by-side offer math
Skills Impact
What each skill adds to pay
Salary Projections
Five-year pay forecasts
Free tools
All Free Tools
Every calculator and generator
Cover Letter Generator
Tailored in one pass
Unemployment Calculator
What you are owed, by state
Salary Widget
Embed salary data anywhere
Resume Score
Grade your resume against a role
Salary Calculator
Base, bonus, equity in minutes
Take-Home Calculator
After federal and state tax
Total Comp Calculator
Full compensation math
Data
Data Catalog
Every role, city, and engine
Companies
54 leveling frameworks
Reports
Compensation Reports
Free summary PDF
International
The US, UK, and Canada
United Kingdom
UK salary data
Canada
Canadian salary data
Trust
Trust Center
How the data is governed
Security
Controls and posture
SLA
Uptime and support commitments
Help
Support
Help center and contact
Compare
Orbyt against the alternatives
Explore Books
The Books
Start reading
Cold Start
Read the opening, free.
Unfair Advantage
Read the opening, free.
The series
Book 1: Cold Start
Reviewing
Book 2: Unfair Advantage
Reviewing
Book 3: Human Heartbeat
Writing
Book 4: Without Me
Future
Book 5: Observer Effect
Future
Explore Blog
The Machine Speaks
Categories
AI Reality
AI-Native
AI Agents
AI Engineering
AI Product
AI Design
AI Strategy
AI Leadership
AI Build
Featured Topics
AI Governance
Agent Organizations
Verification
Human Oversight
Orbyt Collective
AI Safety
AI Economics
All topics
Latest
Obedient Is Not Aligned.
Sep 27, 2026
Twelve seats, no people, and a check that blocks my own commits
Sep 25, 2026
The Model Already Knows TypeScript.
Sep 24, 2026
Half the failures were the instruments, not the code
Sep 23, 2026
The Tenth Man Is on Payroll.
Sep 22, 2026
Explore Pricing
Orbyt Jobs
Orbyt Intelligence
Orbyt One
Plans
Jobs pricing
What each plan includes
All plans
Every product, side by side
Plans
Intelligence pricing
Free, Pro, and Ultra
All plans
Every product, side by side
Billing
All plans
Every product, side by side
Explore Company
About
Who we are
Founder
Justin Bartak, in his own words.
Leadership
One human decides. AI agents advise.
Values
The principles behind the work.
Creed
The company creed.
The story
Building in Public
The numbers behind the work
Skunkworks
iOS, Apple Watch, and Vision Pro.
Contact
Email the team
Orbyt Labs
Products
Research
Developers
Resources
Books
Blog
Pricing
Company
Log inStart
Products
Orbyt JobsOrbyt IntelligenceOrbyt One
Explore Consulting
Orbyt Consulting
Orbyt Jobs
OverviewFeaturesComparePricingAPISalaries
By your situation
Job Search TracksFor Recruiters
Orbyt Intelligence
OverviewFeaturesComparePricingAPI
Start without a card
PlaygroundMCP server
Orbyt One
OverviewPricing
Research
Orbyt Collective
Orbyt Collective
OverviewHow It WorksProcessLeadershipAgent SeatsAutonomy LedgerPulseArticles
Hub
Research Hub
Reference
AI Glossary
Papers
Agent-Native Dataset DesignGoverning an Agent Leadership Team
Field Notes
Every Guard Must Stay QuietThe Cache-Read TaxConfidently WrongWhat an Agent Seat CompletesA Model Upgrade on a Frozen ReviewA Single Agent and a Team
Developers
Orbyt JobsOrbyt IntelligenceOrbyt One
Build
Jobs API Docs
MCP Integrations
Claude DesktopChatGPT GPT ActionsApple ShortcutsZapier / Make.com / n8nOpenClaw
Across products
Developer HubOrbyt API
Build
Intelligence APIWebhooksCLIAPI Changelog
Try
MCP ServerPlaygroundTry It Live
Reference
ReferenceMethodologyEnginesDatasetGlossaryStatus
Resources
Orbyt JobsOrbyt IntelligenceOrbyt One
Learn
Interview PrepAI Skills LabCareer GuidesJob Search ArticlesJob BoardArcade
Salary data
Salary ExplorerAI Role SalariesCitiesIndustriesCompare SalariesCompare OffersSkills ImpactSalary Projections
Free tools
All Free ToolsCover Letter GeneratorUnemployment CalculatorSalary WidgetResume ScoreSalary CalculatorTake-Home CalculatorTotal Comp Calculator
Data
Data CatalogCompanies
Reports
Compensation ReportsInternationalUnited KingdomCanada
Trust
Trust CenterSecuritySLA
Help
SupportCompare
Calculators and tools
Free ToolsSalary CalculatorTake-Home CalculatorTotal Comp CalculatorCompare OffersSkills ImpactSalary Projections 2030Resume ScoreCover Letter GeneratorSalary WidgetUnemployment CalculatorAI Skills Assessment
Books
The Books
Start reading
Cold StartUnfair Advantage
The series
Book 1: Cold StartBook 2: Unfair AdvantageBook 3: Human HeartbeatBook 4: Without MeBook 5: Observer Effect
Blog
The Machine Speaks
Categories
AI RealityAI-NativeAI AgentsAI EngineeringAI ProductAI DesignAI StrategyAI LeadershipAI Build
Featured Topics
AI GovernanceAgent OrganizationsVerificationHuman OversightOrbyt CollectiveAI SafetyAI EconomicsAll topics
Latest
Obedient Is Not Aligned.Twelve seats, no people, and a check that blocks my own commitsThe Model Already Knows TypeScript.Half the failures were the instruments, not the codeThe Tenth Man Is on Payroll.
Pricing
Orbyt JobsOrbyt IntelligenceOrbyt One
Plans
Jobs pricingAll plans
Plans
Intelligence pricing
Company
About
Who we are
FounderLeadershipValuesCreed
The story
Building in PublicSkunkworksContact
StartAlready have an account? Log in
  1. Home/
  2. Orbyt Collective/
  3. Glossary
OverviewHow it worksProcessLeadershipSeatsAutonomyPulseArticlesGlossaryReceipts

Orbyt Collective · Reference

AI Glossary.

The words you meet reading about AI, and the words this company runs on. In plain English.

58 AI terms and 20 Orbyt Collective terms. First edition, Sep, 25 2026. A term added since is marked New for 30 days.

ABCDEFGHIJKLMOPRST
AI terms (58)Orbyt Collective terms (20)

AI terms

58 terms

The field’s vocabulary, A to Z. Each definition links the paper, standard or documentation it rests on.

Agent harnessAlso called agent scaffolding

The software around a model that turns it into an agent: it runs the model step after step and hands it its tools and context. In Orbyt Collective, the word harness alone means something else: the set of guards and checks.

See also AI agent, Tool use, Context engineering, Harness.

Source: Anthropic, Effective harnesses for long-running agents (2025)

AI agentAlso called agent, agentic AI

An AI system that works toward a goal by taking actions with tools (such as searching, running code or editing files) and choosing each next step based on the results. Software built to work this way is called agentic.

In the Machine: Every agent seat that runs this company, with its status and standing.

See also Tool use, Agent harness, Multi-agent system, Autonomy, Agent seat.

Source: Anthropic, Building effective agents (2024)

AI safety

The work of keeping AI systems from causing harm, whether through misuse, accidents or behavior that departs from what people intended. It spans research, testing before and after deployment, and public policy.

See also Alignment, Red teaming, Interpretability, Guardrail.

Source: NIST, AI Risk Management Framework (AI RMF 1.0, 2023)

Alignment

Making AI systems behave in line with human intentions and values, including in situations their makers did not foresee. Researchers disagree about how hard this is and how to tell when it has been achieved.

See also AI safety, Reinforcement learning from human feedback, Reward hacking, Sycophancy, Interpretability.

Source: Ji et al., AI Alignment: A Comprehensive Survey (2023)

Artificial general intelligenceAlso called AGI

AI that could match people across most kinds of intellectual work, rather than excelling at one task. There is no agreed definition or test for it, and researchers disagree about how close current systems are.

See also Superintelligence, Recursive self-improvement, Foundation model.

Source: Morris et al., Levels of AGI for Operationalizing Progress on the Path to AGI (2023)

Autonomy

How much an AI system does without a person approving each step. It comes in degrees, from suggesting an action to taking it and reporting afterward.

In the Machine: Every seat's runs, completions, failures and discards, published.

See also Human in the loop, AI agent, Kill switch, Autonomy Ledger.

Source: Anthropic, Building effective agents (2024)

Benchmark

A defined set of tasks and scoring rules used to measure and compare AI systems. A high score shows skill on those tasks, not necessarily on others.

See also Evaluation.

Source: Liang et al., Holistic Evaluation of Language Models (2022)

Chain of thoughtAlso called CoT

The step-by-step reasoning a model writes out before its final answer. A 2022 Google paper showed that prompting large models for these steps improves their results on arithmetic, commonsense and symbolic reasoning.

See also Reasoning model, Test-time compute, Prompt engineering.

Source: Wei et al., Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (2022)

Coding agent

An AI agent built for software work: it reads a code project, edits files, runs commands and tests, and attempts to fix what fails.

See also AI agent, Agent harness, Sandbox.

Source: Yang et al., SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering (2024)

Compute

The processing power used to train and run AI models, often counted in hours on chips such as GPUs. Research on language models has found that their performance improves predictably as the compute spent training them grows.

See also GPU, Inference, Pretraining, Test-time compute.

Source: Kaplan et al., Scaling Laws for Neural Language Models (2020)

Computer use

The ability of an AI model to operate a computer the way a person does, by looking at the screen, moving the pointer, clicking and typing. Anthropic released it for Claude in public beta in October 2024.

See also AI agent, Tool use, Sandbox.

Source: Anthropic, Introducing computer use (2024)

Context engineering

Deciding what goes into a model's context window at each step (instructions, files, tool results, memory) so an agent has what it needs and nothing that distracts it. Anthropic describes it as the natural progression of prompt engineering.

See also Context window, Prompt engineering, Agent harness, Subagent.

Source: Anthropic, Effective context engineering for AI agents (2025)

Context window

The amount of input and output a model can consider at one time, counted in tokens: the conversation so far, any documents, images or other inputs, and the reply it is writing. Anything outside the window is not available to the model unless it is brought back in.

See also Token, Context engineering, Subagent.

Source: Claude Platform documentation, Context windows

DistillationAlso called knowledge distillation, model distillation

Training a smaller model to reproduce a larger model's outputs, aiming to keep much of the larger one's ability at a fraction of the cost to run.

See also Parameters, Fine-tuning, Inference.

Source: Hinton, Vinyals and Dean, Distilling the Knowledge in a Neural Network (2015)

EmbeddingAlso called vector embedding, embedding vector

A list of numbers that represents a piece of text, an image or other data, learned so that similar items get nearby lists. Embeddings let software compare meaning with arithmetic, which powers semantic search.

See also Semantic search, Retrieval-augmented generation.

Source: Google, Machine Learning Crash Course: Embeddings

EvaluationAlso called eval, evals

A test that measures how well a model or an AI system does a particular job, often run again after each change to catch anything that got worse. A public benchmark is one kind; many evals are built for one product or one task.

See also Benchmark, Red teaming.

Source: Claude Platform documentation, Define success criteria and build evaluations

Fine-tuning

Further training of an already trained model on a smaller, focused set of examples so it gets better at a particular task or style. It changes the model's parameters, whereas a prompt only changes its input.

See also Pretraining, Reinforcement learning from human feedback, Distillation, Parameters.

Source: Google, Machine Learning Glossary: fine-tuning

Foundation model

A large model trained on broad data that can be adapted to many different tasks, like the models behind chat assistants. Researchers at Stanford introduced the term in a 2021 report.

See also Large language model, Pretraining, Fine-tuning, Multimodal.

Source: Bommasani et al., On the Opportunities and Risks of Foundation Models (2021)

Generative AIAlso called GenAI

AI that produces new content, such as text, images, audio, video or code, modeled on the data it learned from and usually in response to a prompt. Chat assistants and image generators are common examples.

See also Large language model, Multimodal, Foundation model.

Source: NIST AI 600-1, Generative Artificial Intelligence Profile (2024)

GPUAlso called graphics processing unit

A chip first built to draw graphics, whose ability to do many calculations at once made it widely used for training and running AI models. In 2012 researchers trained the AlexNet image classifier on two GPUs, an early example of deep learning on graphics chips.

See also Compute, Inference, Pretraining.

Source: Krizhevsky, Sutskever and Hinton, ImageNet Classification with Deep Convolutional Neural Networks (2012)

Grounding

Tying a model's answer to specific sources it was given, such as documents or search results, so the answer can be checked against them. A grounded answer often cites where each claim came from, though a citation alone does not prove the source supports the claim.

See also Retrieval-augmented generation, Hallucination, Semantic search.

Source: Google Cloud documentation, Grounding overview

Guardrail

A rule or check around a model that blocks or corrects unwanted input or output, such as a filter for unsafe content or a limit on which actions an agent may take.

In the Machine: Every place a seat's work can be refused before it lands.

See also AI safety, Sandbox, Human in the loop, Authority matrix.

Source: Rebedea et al., NeMo Guardrails: A Toolkit for Controllable and Safe LLM Applications with Programmable Rails (2023)

Hallucination

When a model confidently states something false or made up, such as a citation, a quote or a number that does not exist. Grounding a model in real sources can reduce it but does not remove it.

See also Grounding, Retrieval-augmented generation, Sycophancy.

Source: Ji et al., Survey of Hallucination in Natural Language Generation (2022)

Human in the loopAlso called HITL

A setup in which a person reviews, corrects or approves an AI system's work at chosen points, during its training or before its output takes effect. Where the person sits, and what they must approve, differs from system to system.

In the Machine: The table of what each seat may do alone and what the human decides.

See also Autonomy, Guardrail, Kill switch.

Source: Wu et al., A Survey of Human-in-the-loop for Machine Learning (2021)

Inference

Running a trained model to produce an output, as opposed to training it. Every reply from a chat assistant is produced by inference.

See also Compute, Test-time compute, Pretraining.

Source: Google, Machine Learning Glossary: inference

Interpretability

Research into what happens inside a model, aiming to explain how it gets from its input to its output. Mechanistic interpretability, one branch of it, tries to identify the specific internal features behind what a model does.

See also Alignment, AI safety, Parameters.

Source: Anthropic, Mapping the mind of a large language model (2024)

Jailbreak

A prompt crafted to get a model to ignore its safety training and produce something it was trained to refuse.

See also Prompt injection, Red teaming, Guardrail.

Source: Wei, Haghtalab and Steinhardt, Jailbroken: How Does LLM Safety Training Fail? (2023)

Knowledge cutoff

The date through which a model's built-in knowledge is expected to be reliable, set by when its training data was gathered.

See also Pretraining, Retrieval-augmented generation, Grounding.

Source: Claude Platform documentation, Models overview (reliable knowledge cutoff)

Large language modelAlso called LLM

A language model with a very large number of parameters, trained on large amounts of text, and capable of a wide range of language tasks.

See also Transformer, Foundation model, Token, Pretraining.

Source: Claude documentation, Glossary: LLM

Long-horizon taskAlso called long-horizon

A task that takes many dependent steps and a long time to finish, such as building a feature or carrying out an investigation. The research group METR measures an agent's progress by how long the tasks it can complete, at a set success rate, take a skilled person.

See also AI agent, Context engineering, Agent harness.

Source: METR, Measuring AI Ability to Complete Long Software Tasks (2025)

Model card

A document published with a model that describes what it is for, how it was evaluated, and its known limits. The format was proposed in a 2018 paper.

See also Evaluation, Red teaming, AI safety.

Source: Mitchell et al., Model Cards for Model Reporting (2018)

Model Context ProtocolAlso called MCP

An open standard for connecting AI applications to outside tools and data, so one integration works with any AI application that supports it. Anthropic introduced it in November 2024.

See also Tool use, AI agent.

Source: Anthropic, Introducing the Model Context Protocol (2024)

Multi-agent systemAlso called multi-agent

A system in which several AI agents act together, each with its own role, instructions or tools. The agents may cooperate on a shared task, debate one another or compete.

In the Machine: The officer seats that run Orbyt Labs, each an agent with its own charter.

See also Orchestration, Subagent, AI agent.

Source: Guo et al., Large Language Model based Multi-Agents: A Survey of Progress and Challenges (2024)

Multimodal

Able to take in or produce more than one kind of data, such as text, images, audio and video. A multimodal model can, for example, describe a photo or answer a question about a chart.

See also Generative AI, Foundation model.

Source: Gemini Team, Gemini: A Family of Highly Capable Multimodal Models (2023)

Open weightsAlso called open-weight model

A model whose trained parameters are published for anyone to download and run. Open weights are not the same as open source: the Open Source Initiative's definition also asks for the code and detailed information about the training data.

See also Parameters, Foundation model, Distillation.

Source: Open Source Initiative, The Open Source AI Definition 1.0

Orchestration

Coordinating several models, agents or tools so they work as one system: deciding who does which step, in what order, and what gets passed between them.

See also Multi-agent system, Subagent, Agent harness.

Source: Anthropic, Building effective agents (2024)

ParametersAlso called weights, model weights

The numbers inside a model, its weights and biases, that are set during training and together hold what it learned. Model size is usually given as a parameter count, such as seven billion.

See also Pretraining, Fine-tuning, Open weights.

Source: Google, Machine Learning Glossary: parameter

PretrainingAlso called pre-training

The first stage of training, in which a model learns general patterns from a large, broad dataset before it is adapted to particular tasks. For language models a common objective is predicting the next token, and later stages, such as fine-tuning and RLHF, shape it into a useful assistant.

See also Fine-tuning, Knowledge cutoff, Compute, Foundation model.

Source: Claude documentation, Glossary: pretraining

Prompt

The input given to an AI model, usually written instructions or a question, together with any text, files or images attached to it.

See also Prompt engineering, System prompt, Context window.

Source: Google, Machine Learning Glossary: prompt

Prompt engineering

Writing and refining prompts to get better and more reliable results from a model, for example by giving clear instructions, examples or a required format.

See also Prompt, Context engineering, System prompt, Chain of thought.

Source: Claude Platform documentation, Prompting best practices

Prompt injection

An attack in which input makes a model behave in ways its builders did not intend, either typed directly or hidden in content the model reads, such as a web page, an email or a document. OWASP's 2025 list ranks it first among the security risks for applications built on language models.

See also Jailbreak, AI agent, Guardrail, Sandbox.

Source: OWASP, LLM01:2025 Prompt Injection

Reasoning model

A model trained to work through a problem in steps before it answers, spending more test-time compute on problems such as math, coding and analysis. Reinforcement learning is one way such models are trained.

See also Chain of thought, Test-time compute, Reinforcement learning.

Source: DeepSeek-AI, DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (2025)

Recursive self-improvementAlso called RSI

A hypothetical process in which an AI system improves its own design, and each improved version is better at making the next improvement. The idea goes back to the statistician Irving John Good, who called the result an intelligence explosion.

See also Superintelligence, Artificial general intelligence, Alignment.

Source: I. J. Good, Speculations Concerning the First Ultraintelligent Machine, Advances in Computers, volume 6

Red teamingAlso called red team

Deliberately attacking an AI system the way an adversary would, to find harmful outputs, security holes or failures, before or after it is released.

See also AI safety, Jailbreak, Prompt injection, Evaluation.

Source: Ganguli et al., Red Teaming Language Models to Reduce Harms (2022)

Reinforcement learningAlso called RL

Training in which a system learns by trial and error: it takes actions in an environment, receives rewards or penalties, and adjusts to earn more reward over time.

See also Reinforcement learning from human feedback, Reward hacking, Reasoning model.

Source: Google, Machine Learning Glossary: reinforcement learning (RL)

Reinforcement learning from human feedbackAlso called RLHF

A training method in which people rank a model's answers, a separate reward model learns to predict their preferences, and the main model is then tuned with reinforcement learning toward the answers people prefer. OpenAI used it in 2022 to make its InstructGPT models better at following instructions.

See also Reinforcement learning, Alignment, Sycophancy, Fine-tuning.

Source: Ouyang et al., Training language models to follow instructions with human feedback (2022)

Retrieval-augmented generationAlso called RAG

A technique in which a system first searches a collection of documents for passages relevant to a question, then gives them to the model so its answer draws on those sources rather than on memory alone. The name comes from a 2020 paper led by researchers at Facebook AI.

See also Grounding, Semantic search, Embedding, Hallucination.

Source: Lewis et al., Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020)

Reward hackingAlso called specification gaming

When a model trained with rewards finds a way to score well that its designers did not intend, such as gaming a test instead of solving the task. It shows that a reward is only a stand-in for what people actually want.

See also Reinforcement learning, Alignment, Sycophancy.

Source: Amodei et al., Concrete Problems in AI Safety (2016)

Sandbox

An isolated environment in which an agent can run code or use tools with limited access to files, networks and other systems, as far as its configuration allows.

See also Guardrail, Coding agent, Prompt injection, Lane.

Source: Claude Code documentation, Sandboxing

Semantic search

Search that finds results by meaning rather than by exact words, often by comparing embeddings, so a search for cheap flights can find a page about low airfares.

See also Embedding, Retrieval-augmented generation.

Source: Reimers and Gurevych, Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks (2019)

Subagent

An AI agent that another agent starts to handle one part of a task. In some tools, such as Claude Code, each subagent works in its own context window and returns its result to the agent that started it.

See also Multi-agent system, Orchestration, Context window.

Source: Claude Code documentation, Subagents

Superintelligence

An intellect that would far exceed the best human minds in practically every field. People disagree about whether and when an AI system could reach it, and about how it could be kept safe.

See also Artificial general intelligence, Recursive self-improvement, Alignment.

Source: Nick Bostrom, How Long Before Superintelligence? (1998)

Sycophancy

A model's tendency to tell people what they want to hear, agreeing with or flattering them even when they are wrong. Research from Anthropic traced part of it to training on human ratings, which tend to favor answers that agree with the person asking.

See also Reinforcement learning from human feedback, Alignment, Hallucination.

Source: Sharma et al., Towards Understanding Sycophancy in Language Models (2023)

System prompt

Instructions given to a model ahead of a conversation, usually by the maker of the app, that set its role, rules and tone. They are usually not shown to the person chatting, though that is the app's choice rather than a guarantee.

See also Prompt, Prompt engineering, Context window.

Source: Claude Platform documentation, Prompting best practices (system prompts)

Test-time computeAlso called inference-time compute

Computation a model spends while answering rather than during training, for example reasoning at length, sampling several answers or searching among them. Research has found that spending it well can improve results on hard problems.

See also Reasoning model, Chain of thought, Inference, Compute.

Source: Snell et al., Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters (2024)

Token

The small unit of text a language model reads and writes, often a whole word or a piece of one. A text model's context window and much of its usage are measured in tokens.

See also Context window, Large language model.

Source: Claude Platform documentation, Glossary: tokens

Tool useAlso called tool call, function calling

A model's ability to use outside tools (such as a search engine, a calculator, a database or running code) by writing a structured request that the surrounding software carries out and whose result the model reads. Each such request is a tool call.

See also AI agent, Model Context Protocol, Agent harness.

Source: Claude Platform documentation, Tool use with Claude

Transformer

A neural network design built around attention, which lets a model weigh each part of its input against other parts; in a model that generates text, each position attends only to what came before it. It was introduced in the 2017 Google paper Attention Is All You Need, and many of today's large language models are built on it.

See also Large language model, Parameters, Pretraining.

Source: Vaswani et al., Attention Is All You Need (2017)

Orbyt Collective terms

20 terms

The words Orbyt Collective uses in its own sense. Each definition is written from the code and records that run the Collective, and links the page where the term is used.

Agent seatAlso called seat

A named role in the org with a written charter, a lane of files it may write, and a line it cannot cross without a human. Most seats are run by an AI agent; a mechanical seat, such as the Inspector General, runs a script with no model.

Used on How it works.

Authority matrix

The committed table of every seat's autonomous and gated lanes, enforced by the workflows that run the org rather than by instructions to the model.

Used on How it works.

Autonomy Ledger

Run, completion, failure and discard totals by seat, plus a frozen baseline.

Used on How it works.

Completion gate

The check that decides, when an agent's run ends, whether its work may go on to the lane check and the guards. A run that exited with an error or ended in one lands nothing, a run that says it had nothing to do ends the cycle quietly, and any other run goes on if it printed its completion line or, without that line, if its exit, its result and its turns against its cap show it healthy.

Used on How it works.

Dead-man switch

The reverse brake: if fourteen days pass with no signed commit under the founder's name, promotions pause where they are.

Used on How it works.

Decision Ledger

A JSON record of structural decisions: ids, dates, supersession links and reviewed titles. CC BY 4.0.

Used on How it works.

Decision log

The append-only record of structural decisions, each with a permanent id. Corrections are new entries that name what they supersede.

Used on How it works.

Discard

A file an agent changed that its run may not keep, erased before the commit rather than shipped: anything outside its lane, and a lane file the run cannot keep yet, such as a proposed lesson before the ladder allows one. Counted separately from failures, because a run that landed with some of its edits refused is a different fact from a run that never landed. It is an aggregate counter, not a count of distinct files: the advisory panel isolates once for all observers and writes the same panel-wide count into every observer's row. No panel judges a discard.

Used on How it works.

Evaluator

Two scripts with no AI in them that judge the org by fixed rules from its committed records. One sets each seat's standing from the record of its runs, and one promotes or rolls back the ladder's stages.

Used on Agent Seats.

Failure Corpus

Reviewed failure records and paths to recorded countermeasures in the private repository.

Used on How it works.

Harness

The Collective's audit harness: its guard scripts and its audit dimensions (each one a kind of failure that has really happened). Agents may add dimensions that only report, but turning one into a gate that can block work takes a human signature.

Used on Process.

Kill switchAlso called halt

One committed file that stops every autonomous seat before it spends anything, checked ahead of the credential.

Used on How it works.

Ladder

The staged rollout of new machinery in the org. A script with no AI in it holds, promotes or rolls back the active stage on committed evidence wherever that stage's rules are written, and pauses promotions if the founder is silent for fourteen days.

Used on Process.

Lane

The set of files a seat may write, where anything changed outside it is counted and erased before the commit. On the authority matrix, a lane is also one kind of work (autonomous or gated).

Used on How it works.

Lenses

The roster of perspectives each seat must hold its draft against before calling it done: mostly the published work of named real people, plus one lens that is a function rather than a person. The roster is the company's own reading of that work, and none of the people on it endorses it.

Used on Leadership.

Orbyt Collective

The agent leadership team that runs Orbyt Labs, built as a separate package that names no company, so it could be installed at another one.

Used on the Orbyt Collective overview.

Pulse

The org's daily report on itself, written by a script with no AI in it. It gives a verdict for each thing it watches, and says so when it could not look.

Used on Pulse.

Receipts

The public records a claim about the Collective can be checked against: the Autonomy Ledger, the Decision Ledger and the Failure Corpus. Each is generated from private records, the full decision log, the run rows and the commits, which are not published.

Used on How it works.

SentinelAlso called completion line

The completion line a seat prints to say it finished, recorded on the run's row as the agent's claim. A healthy run goes on without it, and with or without it the lane check and the guards decide what lands.

Used on Process.

The Machine

The name for Orbyt Collective at work: the agent leadership team that runs Orbyt Labs. It advises, drafts, builds and watches, and one human decides.

Used on the Orbyt Collective overview.

Common questions.

Two sets of words. The AI terms are the field's vocabulary, the words you meet reading about AI today. The Orbyt Collective terms are the words this company's agent organization uses in its own sense.

Each AI definition is written in plain English and links the paper, standard or maker's documentation it rests on. Each Orbyt Collective definition is written from the code and records that run the Collective, and links the page on this site where the term is used. Where that page and the code disagree, the definition follows the code.

Each term records when it was added and when it last changed, and a changed term shows its date. A term added after the first edition is marked New for 30 days.

AI agent is the field's term: an AI system that works toward a goal by taking actions with tools (such as searching, running code or editing files) and choosing each next step based on the results. Software built to work this way is called agentic. Agent seat is Orbyt Collective's own term: a named role in the org with a written charter, a lane of files it may write, and a line it cannot cross without a human. Most seats are run by an AI agent; a mechanical seat, such as the Inspector General, runs a script with no model.

Collective

  • Explore Collective
  • How it works
  • Process
  • Leadership
  • Agent Seats
  • Autonomy Ledger
  • Pulse
  • Articles
  • Influences
  • Glossary

In the skunkworks

  • Skunkworks

More from Orbyt

  • Orbyt Jobs
  • Orbyt Intelligence
  • Orbyt One
  • About Orbyt Labs

Keep Exploring

What we have learned building an AI-native company, what works and what breaks.

Products

  • Orbyt Jobs
  • Orbyt Intelligence
  • Orbyt One
  • Orbyt Consulting

Research

  • Orbyt Collective

Developers

  • Orbyt API & MCP
  • Jobs API
  • Intelligence API

Publishing

  • Books
  • Blog
  • Papers

Help

  • Support
  • Contact
  • Status

Company

  • Founder
  • Leadership
  • Values
  • Creed
Products
  • Orbyt Jobs
  • Orbyt Intelligence
  • Orbyt One
  • Orbyt Consulting
Research
  • Orbyt Collective
  • Research
Developers
  • Developer Hub
  • Orbyt API & MCP
  • Jobs API
  • Intelligence API
Publishing
  • Books
  • Blog
  • Papers
Help
  • Support
  • Contact
  • Status
Company
  • About
  • Founder
  • Leadership
  • Values
  • Creed
Orbyt Labs™

© 2026 Purecraft LLC  All rights reserved.

Privacy·Terms·Security·Trademark·Accessibility·DPA·Refund·Status·Sitemap

Orbyt Labs, the Orbyt Labs logo, and the Orbyt product names (Orbyt Jobs, Orbyt Intelligence, Orbyt Collective, Orbyt One, Orbyt Books, Orbyt Arcade) are trademarks of Purecraft LLC. Product names, logos, and brands of others are the property of their respective owners. Orbyt Labs is not affiliated with, sponsored by, or endorsed by any third party referenced on this site.