Knowledge Base

Clear explanations of the ideas, methods and technologies shaping AI Native software delivery.

Stay up to date

Every month, we share the latest additions and updates from the Knowledge Base.

131 terms · growing with what you tell us is missing

A

AbsorptionnewAbsorption is an organisation's capacity to take in change: its customers, staff, support teams and regulators.1 min

Activation functionA mathematical rule inside an AI model that decides whether a piece of information is passed along to the next part of the model.1 min

Agent and factory platformnewA shared internal platform giving teams context, tools, environments, permissions and validation for AI agents as self-service.1 min

Agentic codingAgentic coding delegates a whole task to an AI agent that plans, edits, runs tests and iterates until it is done.21 min

AGENTS.mdAGENTS.md is a Markdown file in a repository that gives AI coding agents project-specific instructions. Which tools read it, and which do not.3 min

AI agent evaluationAI agent evaluation measures whether an agent does its job, scoring the path it took as well as the answer, across repeated runs.3 min

AI agentsAn autonomous program that understands a goal, makes a plan and carries out a series of tasks with software tools to achieve it.2 min

AI biasBias in an AI system is a systematic error that produces unfair or discriminatory outcomes against certain groups.2 min

AI guardrailsRules and constraints that keep an AI system operating safely, ethically and within its intended purpose.3 min

AI hallucinationAn AI hallucination is when a generative AI model produces false, nonsensical, or factually incorrect information but presents it as if it were true.2 min

AI modelAn AI model is a specialised digital file that has been trained on data to recognise patterns and make predictions or decisions.2 min

AI NativenewAI Native describes organisations, teams and systems designed around the assumption that machine execution is cheap, fast, parallel and capable.13 min

AI native managementnewMission leadership, people leadership and system management: the three responsibilities that management holds in an AI native organisation.1 min

AI native operating modelnewThe system by which an organisation turns intent into outcomes using people and machines together, deciding where machines act and people judge.1 min

AI productivity paradoxThe gap between AI making individual developers faster and organisational delivery not improving at all.2 min

AlgorithmAn algorithm is a set of step-by-step instructions a computer follows to accomplish a specific task.1 min

Artificial general intelligenceA theoretical form of AI that could understand, learn and apply its intelligence to any intellectual problem a human being can.2 min

Artificial intelligence (AI)The field of computer science that builds systems able to do tasks that normally need human intelligence, such as learning and reasoning.1 min

Artificial neural networkA computing system inspired by the human brain that learns to find patterns and make decisions from large amounts of data.2 min

AttentionA mechanism that lets an AI model focus on the most relevant parts of its input when making a decision or generating output.2 min

Attention budgetnewThe number of decisions and reviews a person can handle well in a given period, treated as a limit the system must respect.1 min

Automation follows trustnewA principle that a check becomes automatic only once the signal behind it has proved reliable.1 min

Autonomy enabling teamnewA team that helps other teams delegate safely to AI agents by paying down the system debt that keeps people in the loop unnecessarily.1 min

B

Build loopnewThe part of delivery that runs from a specification to a merged, tested change.1 min

C

Chain-of-thoughtChain-of-thought is a method for instructing an AI to break down a complex problem into a series of logical steps before giving a final answer.2 min

ChatbotA chatbot is a computer program that simulates human conversation through text or voice commands.2 min

Comprehension checknewA short probe, taken before a consequential decision, to find out whether the person deciding understood what changed and what follows.1 min

Comprehension debtnewThe growing gap between how fast AI-assisted work changes and how well the people responsible for it understand it.1 min

Comprehension deltanewAn explanation of what a change means for a specific person's understanding, relative to what they last knew.1 min

Comprehension engineeringnewDesigning the information, interactions and feedback that give a person enough understanding, at the right moment, for the judgment they must make.1 min

Computer visionThe field of AI that enables computers to derive meaningful information from images, videos and other visual inputs.1 min

Confidence scoreA confidence score is a number, typically a percentage, that represents how certain an AI model is about its own prediction or decision.2 min

Constraint migrationnewThe way the bottleneck in delivery moves to a new place each time AI speeds up an earlier step.1 min

Context engineeringnewGiving AI models and agents the right information, in the right form, at the right time, so they can act well.1 min

Context windowThe context window is the amount of text an AI model can remember and consider at one time when generating a response.1 min

Contraction triggernewA condition, written down in advance, that narrows what an AI agent is allowed to do.1 min

Control RoomnewWhere an organisation using AI at scale sees where judgment is needed, where work is waiting, what evidence exists and what risk is changing.1 min

Cross-entropyA training measure that scores how wrong an AI model's predictions are, penalizing confident mistakes most heavily.1 min

D

Dark factorynewA software factory that runs with very little routine human involvement.1 min

Data miningData mining is the process of automatically discovering patterns, connections, and valuable insights from large sets of business data.2 min

Data scienceData science is the practice of turning raw business data into actionable insights and predictions that can guide strategic decisions.1 min

Decision queuenewA list of the judgments an AI-enabled system could not make on its own, each prepared well enough for a person to make it well.1 min

Decision readinessnewHaving enough understanding of the current situation to make a particular decision responsibly.1 min

Decision, Delta, ChecknewThe three-part model of comprehension engineering: the judgment needed, what changed in meaning, and whether understanding arrived.1 min

Deep learningA type of AI that learns from vast amounts of data using layered neural networks loosely modeled on the human brain.2 min

Delegation boundarynewThe line between what AI agents may do on their own and what still needs a person, and the rules for how that line moves.1 min

Delivery loopnewThe part of delivery that runs from an idea to a measured outcome, spanning product, engineering and operations — and where most of the waiting is.1 min

DORA metricsThe five measures of software delivery performance: definitions, 2024–25 benchmarks, how to measure them, and what AI changes.21 min

E

EmbeddingsA way to turn text, images or user profiles into numbers that computers can easily compare and analyse.1 min

Emergent behaviorCapabilities or characteristics that arise unexpectedly from the interactions inside an AI system, without being explicitly programmed.1 min

Enterprise mapnewA shared, navigable model of an organisation's missions, people, agents, decisions, risks, evidence and outcomes.1 min

Enterprise steeringnewThe function that owns decision rights across missions, allocates scarce human attention and resolves conflicts between missions.1 min

Entity extractionEntity extraction is an AI technique that automatically identifies and classifies key pieces of information within text into predefined categories.2 min

Explainable AI (XAI)Tools and methods that let people understand and trust the results produced by artificial intelligence models.2 min

F

Fan-outnewHow many streams of work one person can effectively look after, set by how long the work stays safe unattended and how long each engagement costs.1 min

Farm and construction sitenewAn analogy for which work becomes highly automated and which still needs many people, and why the two need different delegation settings.1 min

Few-shot learningFew-shot learning is a technique that allows an AI model to learn a new task from just a handful of examples, rather than thousands.2 min

Fine-tuningAdapting a general-purpose AI model so it excels at a specific task or understands a company's own data.2 min

FinOps for AIFinOps for AI is the practice of managing AI spend as an operational discipline, metered in tokens rather than instance-hours.2 min

Foundation modelA very large AI system trained on broad data that can be adapted to perform many different, specific tasks.2 min

Function callingFunction calling is a feature that allows an AI model to use external tools and access live information to answer questions or complete tasks.1 min

G

Generative adversarial networkAn AI system in which two neural networks compete to create new, realistic data such as images, audio and video.2 min

Generative AIGenerative AI is a type of artificial intelligence that can create entirely new and original content, such as text, images, audio, and code.2 min

GroundingGrounding is the process of connecting an AI's responses to specific, verifiable sources of information to ensure accuracy and relevance.1 min

H

Handoff contractnewAn agreement about what must travel with control when work passes between a person and an AI agent.1 min

Human attention per outcomenewA measure of how much human effort one useful outcome consumes, counting everyone involved.1 min

Human oversightnewA state in which a person understands enough to judge what an AI system is doing. Being present in the workflow is not enough.1 min

Hybrid teamnewA team with stable human accountability and machine capacity that grows, shrinks and changes shape as the work changes.1 min

HyperparameterA hyperparameter is a setting a data scientist adjusts before training an AI model to control how it learns.2 min

I

InferenceInference is the process of using a trained AI model to make predictions or decisions based on new, real-world data.1 min

Intelligent document processingAn AI technology that automatically reads, understands and extracts key information from business documents.2 min

Intent recognitionIntent recognition is the AI process of identifying the underlying goal or purpose of a user's typed or spoken request.2 min

InvariantsnewThe properties that must not change, or did not change, when an agent makes a change.1 min

L

LabelA label is a tag that provides the correct answer or classification for a piece of data, used to teach an AI system.2 min

Large language model (LLM)A large language model (LLM) is an artificial intelligence program trained to understand and generate human-like text.2 min

M

Machine learningA type of AI that lets computers learn from data and improve at a task without being explicitly programmed for every outcome.2 min

MissionnewThe unit of work in an AI native organisation: an accountable outcome, written down, with the people, agents, authority and evidence around it.1 min

Mission teamnewThe AI native version of a stream-aligned team: an accountable owner and a small, stable human core, with machine capacity that changes per mission.1 min

Multi-agent systemA system in which multiple independent software agents work together to solve a problem too complex for a single program.2 min

Multimodal AIAI that can understand, process and generate several types of data at once, such as text, images, audio and video.2 min

N

Natural language processing (NLP)The field of AI that gives computers the ability to understand, interpret and generate human language.1 min

Natural language understanding (NLU)A type of AI that enables computers to comprehend human language, including its meaning, intent and context.1 min

Neural networkA type of AI that learns to find complex patterns in data, loosely inspired by the structure of the human brain.2 min

O

OntologyA formal system that creates a shared understanding of a knowledge area by defining its key concepts and how they relate.2 min

Operating loopnewThe top-level model of AI native work: Intent, Delegate, Execute and Judge, with Learn and Adapt at the centre.1 min

Optical character recognition (OCR)Technology that converts images of typed, handwritten or printed text into machine-readable text data.1 min

Outcome checknewA check, declared before work starts, of whether a change achieved what it was meant to.1 min

OverfittingOverfitting occurs when an AI model learns its training data so perfectly that it fails to make accurate predictions on new, real-world data.2 min

P

ParametersThe internal variables an AI model adjusts during training, which together represent what it has learned from data.2 min

Pattern recognitionA core function of AI that automatically identifies regularities, rules or recurring structures within data.1 min

Pre-trainingTraining an AI model on a massive, general dataset to give it broad understanding before it is specialised for a task.2 min

PrecisionPrecision is a measure of an AI model's accuracy, specifically for the predictions it identifies as positive or correct.2 min

Predictive analyticsUsing historical data, statistical algorithms and machine learning to estimate the likelihood of future outcomes.1 min

PromptA prompt is the instruction, question, or piece of information you give to an AI system to get it to perform a specific task.1 min

Prompt engineeringCrafting instructions and questions that guide a generative AI model to produce accurate, relevant and useful output.1 min

R

Re-entry briefnewWhat should be waiting when a person comes back to AI-assisted work that moved on without them.1 min

RecallRecall is a performance metric that measures how effectively an AI model identifies all the truly relevant instances in a dataset.2 min

Reinforcement learningA type of AI that learns the best way to do something through trial and error, guided by rewards and penalties.2 min

Retrieval-augmented generation (RAG)A technique that gives an AI language model specific, external information to draw on before it generates a response.1 min

Robotic process automation (RPA)Technology that uses software robots to automate the repetitive, rule-based tasks people do on computers.2 min

RoboticsRobotics is the field of designing, building, and programming machines that can automatically carry out physical tasks in the real world.2 min

ROC curveA ROC curve is a visual tool used to evaluate the performance of an AI model that makes a "yes or no" decision.2 min

S

Seniority shiftnewThe expected move of seniority from years of building things by hand to skill in guiding machines, combined with the right context.1 min

Sentiment analysisSentiment analysis is the process of using AI to automatically identify and categorise the emotional tone within a piece of writing.2 min

Software factoryThe complete delivery system that turns intent into trusted, released software — and what it becomes once agents arrive.26 min

Spec-driven development (SDD)Building from an executable specification an agent can implement and a pipeline can verify — and why it matters now.22 min

Specialist judgment servicenewSecurity, legal, architecture and domain experts offered as planned decision capacity that missions call on with a prepared brief.1 min

State reconstructionnewA rebuilt picture of the current state of the work, used when a delta can no longer bridge the gap.1 min

Strong AIA theoretical form of AI that could understand, learn and apply its intelligence to solve any problem, just like a human being.2 min

Supervised learningSupervised learning is a type of artificial intelligence where a system learns from data that has been pre-labeled with the correct answers.2 min

System debtnewA human intervention that happened because something in the system was missing, when no real judgment was needed.1 min

System promptA system prompt is a set of initial instructions given to an AI model to guide its behavior, personality, and the rules it must follow.1 min

T

TemperatureTemperature is a setting in AI models that controls the creativity and randomness of its output.1 min

Throughput policenewPeople whose job at a gate is to confirm, again, that something which is nearly always fine is fine.1 min

Time to decision-readynewThe time between a person returning to AI-assisted work and being ready to make the decision it needs.1 min

TokenA token is a fundamental piece of text, like a word or part of a word, that an AI model uses to process and understand information.1 min

Tooling substitutionnewBuying the artefacts of change, such as a portal, a pipeline or AI assistants, instead of changing how work flows.1 min

Top-k samplingA technique that makes AI-generated text more varied by choosing each next word from a short list of the most probable options.2 min

Top-p samplingA technique that controls the randomness of AI-generated text by picking the next word from a dynamically sized pool of likely options.2 min

Training dataTraining data is the collection of examples, such as text, images, or sales figures, used to teach an AI system how to make predictions or decisions.2 min

Transfer learningTransfer learning is a technique for repurposing a pre-existing AI model, trained for one task, to serve as a starting point for a new, related task.1 min

TransformerA Transformer is a powerful AI architecture that excels at understanding context and relationships within sequential data, like human language.1 min

Turing testThe Turing test is a test to determine whether a machine can exhibit intelligent behavior indistinguishable from that of a human.2 min

Two leversnewRemoving human gates, and improving the ones that remain. Lever two is what makes lever one safe to push further.1 min

U

UnderfittingUnderfitting is when an AI model is too simple to capture the underlying patterns in your business data, leading to poor and inaccurate predictions.1 min

Unsupervised learningUnsupervised learning is a type of AI that finds hidden patterns and structures within data without any predefined labels or human guidance.2 min

V

Validation dataValidation data is a sample of data used during an AI model's development to fine-tune its settings and check its performance before final testing.2 min

Vector databaseA database that stores and searches data by meaning and conceptual relationships, not just exact keywords.2 min

W

Weak AIWeak AI, also known as Narrow AI, is an artificial intelligence system designed and trained to perform one specific, focused task.1 min

Z

Zero-shot learningZero-shot learning is the ability of an AI model to recognise or classify things it has never been specifically trained on before.2 min

Something missing?

Tell us the term you expected to find, or correct one we have already written. We credit every contribution we publish.