Jargon Glossary

A practical glossary of tech, software, games, fintech and AI terms

A plain-English guide to the business models, metrics and technical terms that shape today’s software industries.

The technology industry runs on a language of its own. Some of it is useful shorthand. Much of it can be opaque to anyone outside investor updates, strategy meetings, product teams or specialist reporting. This glossary is designed as a map of the modern tech economy, explaining the business, product, technical and sector-specific terms that shape how today’s software industries are discussed.

Technology has always generated specialist language. Some of it emerges because new products and business models need new terms. Some of it comes from finance, software development and management theory. Over time, that language becomes part of how companies are assessed, how sectors are described and how change is explained.

The result is that even straightforward reporting can become crowded with terms that assume a level of familiarity many readers do not have. That is especially true in software, internet platforms, games, fintech and artificial intelligence, where business models, customer economics and infrastructure matter as much as the products themselves.

This glossary is intended as a practical guide to that language. Rather than treating the modern tech economy as one undifferentiated mass of jargon, it groups terms by the parts of the ecosystem they belong to, from strategy and fundraising to growth, engineering, advertising, payments and AI.

Core Business & Strategy

These are the broad concepts that recur across the technology industry, regardless of sector. They are used to describe what makes a business viable, defensible and capable of growth, and they often appear in company strategy, investor conversations and market analysis.

Key terms at a glance

Term	Definition	Why it matters
Product-market fit	Product-market fit describes the point at which a product clearly meets a real customer need, and demand starts to emerge in a sustainable way.	It is one of the clearest signs that a business has moved beyond an idea and is building something people genuinely want.
Total addressable market (TAM)	Total addressable market, usually shortened to TAM, is the overall revenue opportunity available if a company captured every possible customer in its market.	Founders and investors use TAM to judge how large a business could become, although the figure is often more theoretical than realistic.
Moat	A moat is a durable competitive advantage that makes it hard for rivals to copy or displace a business. That advantage might come from brand strength, technology, scale, regulation or customer lock-in.	Companies with strong moats are usually better placed to defend margins and market share over time.
Network effects	Network effects occur when a product becomes more valuable as more people use it. Social platforms, marketplaces and payments networks are common examples.	Network effects can help businesses grow quickly and create barriers to entry for competitors.
Scalability	Scalability refers to a company’s ability to grow revenue or usage without costs rising at the same rate.	Scalable businesses are attractive because growth can translate more efficiently into profit.
Lock-in	Lock-in describes a situation in which customers are unlikely to switch products or suppliers because the costs, friction or disruption of leaving are too high.	Lock-in can make revenue more durable, but it can also create dependency and reduce competitive pressure.
Unit economics	Unit economics refers to the basic financial performance of a business on a per-unit basis, such as per customer, per order or per transaction.	Strong unit economics suggest that growth is creating value rather than simply increasing losses.
Monetization	Monetization is the process by which a company turns a product, service, audience or activity into revenue.	In technology, a product can attract plenty of users without proving it has a viable business model. Monetization is where engagement starts to translate into income.
Go-to-market (GTM)	Go-to-market, usually shortened to GTM, refers to the strategy a company uses to bring a product to customers, including pricing, distribution, sales and marketing.	Even strong products can struggle if the route to market is weak or poorly timed.
Pivot	A pivot is a meaningful change in a company’s strategy, product, target market or business model.	Startups often pivot when their original plan fails to gain traction, although the term is sometimes used too loosely to make a setback sound more deliberate than it was.
Burn rate	Burn rate refers to how quickly a company is spending its cash reserves, usually measured monthly.	For startups and growth-stage firms, burn rate is a key indicator of how long the business can operate before it needs more funding or reaches profitability.
Runway	Runway is the amount of time a company can continue operating before it runs out of cash, based on its current burn rate.	It is one of the most closely watched measures of financial resilience in venture-backed businesses.

Startups, Venture Capital & Fundraising

Technology companies are often shaped as much by ownership and funding structures as by products or revenue growth. This is the language of startup equity, venture capital and fundraising, and it matters because it determines how value, control and risk are distributed as a company grows.

Key terms at a glance

Term	Definition	Why it matters
Vesting	Vesting refers to the process by which an employee, founder or investor earns rights to shares or options over time, rather than receiving them all immediately.	Vesting is widely used to align incentives and encourage people to stay with a company over the longer term.
Cliff	A cliff is the period at the start of an equity vesting schedule during which no shares vest. Once the cliff is reached, a larger first portion typically vests at once.	Cliffs are designed to stop employees from receiving equity immediately if they leave early, and are a standard feature of startup compensation packages.
Option pool	An option pool is a portion of a company’s equity set aside for employees, usually in the form of share options.	Option pools are central to startup hiring and compensation, but they also dilute existing shareholders.
Dilution	Dilution occurs when a shareholder’s ownership percentage falls because the company issues more shares.	Dilution is a routine part of startup fundraising, but it affects founders, employees and early investors differently depending on the terms.
Capitalisation table (cap table)	A capitalisation table, usually shortened to cap table, is the record of who owns what in a company, including founders, employees and investors.	The cap table is one of the clearest ways to understand control, incentives and the financial consequences of future funding rounds.
Preference shares	Preference shares are shares that carry special rights, often giving investors priority over ordinary shareholders when a company is sold, wound up or pays distributions.	Preference shares are a core feature of venture financing and can have a major impact on who receives what in an exit.
Liquidation preference	Liquidation preference is the right of certain investors, usually holders of preference shares, to get their money back before ordinary shareholders receive proceeds from a sale or liquidation.	Liquidation preferences can significantly affect founder and employee payouts, especially if an exit value is lower than headline valuations suggest.
Term sheet	A term sheet is the non-binding document that sets out the main commercial terms of an investment before full legal agreements are drafted.	The term sheet often determines the most important economic and control rights in a funding round.
Simple Agreement for Future Equity (SAFE)	A Simple Agreement for Future Equity, usually shortened to SAFE, is an investment instrument that converts into shares at a later financing round rather than pricing the company immediately.	SAFEs became popular because they are simpler than priced rounds, but they can also create complexity later if too many accumulate.
Convertible note	A convertible note is a form of debt that converts into equity at a later funding round, usually on pre-agreed terms.	Convertible notes can delay valuation negotiations, but they still shape dilution and ownership when they convert.
Pro rata rights	Pro rata rights give an investor the option to maintain their percentage ownership in future funding rounds by investing again.	These rights can be valuable in successful companies because they let early investors preserve their stake as the business grows.
Down round	A down round is a fundraising round in which a company raises capital at a lower valuation than in its previous round.	Down rounds can hurt morale, dilute existing shareholders more aggressively and signal that a company’s growth story has weakened.

Product, Growth & Customer Economics

This is the language of customer acquisition, retention and commercial performance. These terms are especially common in product teams, growth functions and software businesses, but they increasingly shape how the wider tech economy is measured.

Key terms at a glance

Term	Definition	Why it matters
Customer acquisition cost (CAC)	Customer acquisition cost, usually shortened to CAC, is the amount a business spends to acquire a new customer. This usually includes sales and marketing costs.	A high CAC can be sustainable if customers are valuable and sticky, but dangerous if retention is weak.
Lifetime value (LTV)	Lifetime value, usually shortened to LTV, estimates how much revenue or profit a customer will generate over the course of their relationship with a business.	LTV is often compared with CAC to judge whether a company’s growth model is economically sound.
Churn	Churn is the rate at which customers cancel a subscription, stop using a product or otherwise drop out over a given period.	Even fast-growing businesses can struggle if they are losing customers too quickly.
Conversion rate	Conversion rate is the percentage of users who complete a desired action, such as making a purchase, signing up or downloading an app.	It is one of the clearest measures of how effectively a product, page or campaign turns interest into action.
Onboarding	Onboarding refers to the process of helping new users or customers set up, understand and start using a product effectively.	Good onboarding can improve activation, reduce churn and increase the chances that customers see value early.
Freemium	Freemium describes a model in which a basic version of a product is offered for free, while advanced features or higher usage levels are charged for.	Freemium can be an efficient way to drive adoption, but it only works if a meaningful share of users eventually convert to paying plans.
Retention	Retention measures how many users continue to return to a product over time.	Retention is one of the clearest indicators that a product is delivering ongoing value rather than attracting only fleeting interest.
Retention loop	A retention loop is the system of habits, rewards and recurring triggers designed to keep users coming back repeatedly over time.	Strong retention loops are often the difference between a product that is tried once and one that becomes part of a user’s routine.
Customer relationship management (CRM)	Customer relationship management, usually shortened to CRM, refers both to the practice of managing customer interactions and to the software systems used to do it.	CRM systems sit at the centre of many sales, service and marketing operations, especially when companies are trying to build a fuller picture of their customers.
Attribution	Attribution refers to the process of deciding which marketing activity should get credit for a conversion, sale or other desired outcome.	Attribution shapes how companies judge advertising performance, although in practice it is often imperfect and contested.
Single customer view (SCV)	Single customer view, usually shortened to SCV, refers to a consolidated picture of a customer created by combining data from multiple sources, such as a company’s website, mobile app, email campaigns, purchase history and support records.	Companies use SCV to understand customer behaviour more clearly, improve targeting and personalise products or marketing. In practice, building that unified view is often difficult.
Identity resolution	Identity resolution is the process of linking different identifiers, such as email addresses, device IDs and login details, to the same person or household.	It is a core part of customer data strategy, especially when businesses are trying to understand users across devices and channels.

Software & SaaS Business Models

Software businesses have developed their own vocabulary around recurring revenue, pricing and customer expansion. These terms help explain how software companies grow and how investors evaluate them.

Key terms at a glance

Term	Definition	Why it matters
Software-as-a-service (SaaS)	Software-as-a-service, usually shortened to SaaS, refers to software delivered over the internet on a subscription basis rather than sold as a one-off product.	SaaS changed the economics of software by shifting revenue towards recurring customer relationships rather than up-front licence sales.
Annual recurring revenue (ARR)	Annual recurring revenue, usually shortened to ARR, measures the value of subscription revenue a company expects to repeat over a year.	ARR is one of the most widely used indicators of scale and momentum in software businesses.
Monthly recurring revenue (MRR)	Monthly recurring revenue, usually shortened to MRR, is the monthly equivalent of ARR, showing how much predictable subscription revenue a company generates each month.	It is often used by earlier-stage software businesses tracking growth at a more granular level.
Annual contract value (ACV)	Annual contract value, usually shortened to ACV, measures the average annual revenue generated from a customer contract.	ACV is useful for understanding deal size and the kind of customers a software company is selling to.
Net revenue retention (NRR)	Net revenue retention, often shortened to NRR, measures how recurring revenue from an existing customer base changes over time, taking account of upgrades, downgrades and cancellations.	Strong NRR suggests customers are not just staying, but spending more.
Seat-based pricing	Seat-based pricing is a model in which customers pay according to the number of users who have access to a product.	It is one of the most common pricing models in software, especially in workplace tools and enterprise services.
Platform take rate	Platform take rate is the share of revenue retained by a platform owner when a product, service or transaction is sold through its ecosystem.	Take rates have major implications for margins, pricing strategy and the economics of companies that depend on larger platforms for distribution.

Software Development & Engineering

This is the technical language of how software is planned, built, tested and shipped. Some of these terms describe methodologies, while others refer to release practices, code management or software architecture.

Key terms at a glance

Term	Definition	Why it matters
Waterfall	Waterfall is a linear software development approach in which work moves through fixed stages such as planning, design, development, testing and release.	Waterfall can provide clarity and structure, but it is often criticised for being too rigid when requirements change mid-project.
Agile	Agile refers to a family of software development approaches built around iterative delivery, regular feedback and continuous adjustment, rather than long development cycles with fixed requirements.	Agile became influential because it offered a more flexible alternative to rigid planning models, especially in fast-changing software environments.
Scrum	Scrum is a project management framework commonly used within Agile development, built around fixed work cycles, defined team roles and regular review points.	Scrum gave many software teams a practical structure for applying Agile ideas in day-to-day work.
Kanban	Kanban is a workflow management method that focuses on visualising tasks, limiting work in progress and improving flow.	Kanban is often used by teams that want more flexibility than fixed sprint-based planning allows.
Sprint	A sprint is a fixed period of work, often one or two weeks, in which a team aims to complete a defined set of tasks or features.	Sprints are a common feature of Agile working and reflect the broader shift towards iterative planning and delivery.
Backlog	A backlog is the prioritised list of tasks, features, bugs or improvements that a software team plans to work through over time.	The backlog is where strategy, product priorities and engineering capacity meet.
Minimum viable product (MVP)	Minimum viable product, usually shortened to MVP, refers to the simplest version of a product that can be launched in order to test demand or gather feedback.	MVP thinking is meant to reduce wasted effort, although the term is sometimes used to justify shipping something half-finished.
Development and operations (DevOps)	Development and operations, usually shortened to DevOps, refers to a set of practices intended to improve collaboration between software development and IT operations teams, often with an emphasis on automation and continuous delivery.	DevOps is meant to help organisations release software more quickly and reliably, while reducing the friction between building products and running them in production.
Continuous integration / continuous delivery (CI/CD)	Continuous integration / continuous delivery, often shortened to CI/CD, refers to practices in which code changes are regularly merged, tested and prepared for release through automated pipelines.	CI/CD helps software teams ship updates more frequently and spot problems earlier in the development cycle.
Deployment	Deployment is the process of releasing software changes into a live environment where users can access them.	How often and how safely a company can deploy is one of the clearest indicators of engineering maturity.
Staging environment	A staging environment is a testing environment that closely mirrors the live product, allowing teams to check changes before release.	Staging helps reduce the risk of introducing problems into production.
Rollback	A rollback is the process of reversing a software release after something goes wrong.	The ability to roll back quickly is an important safeguard when companies are shipping changes frequently.
Version control	Version control is the system used to track changes to code over time, making it possible for multiple developers to collaborate without overwriting each other’s work.	It is foundational to modern software development because it supports collaboration, accountability and recovery when something breaks.
Repository (repo)	A repository, often shortened to repo, is the storage location where code and its version history are kept.	Repositories are basic infrastructure for collaborative development and code management.
Pull request (PR)	A pull request, usually shortened to PR, is a proposed change to code that is submitted for review before being merged into the main codebase.	Pull requests make collaboration more structured by creating a formal point for review and discussion.
Application programming interface (API)	An application programming interface, usually shortened to API, is a set of rules and tools that allows different software systems to communicate with each other.	APIs are central to how modern software is built, connected and extended, especially in cloud services and platform businesses.
Stack	A stack refers to the combination of technologies used to build and run a software product, such as programming languages, databases, frameworks and hosting tools.	A company’s stack influences performance, hiring, maintenance and the pace at which new features can be delivered.
Technical debt	Technical debt describes the future cost created when software is built quickly or inelegantly in ways that will need to be fixed later.	Technical debt can help teams move faster in the short term, but it often slows development and raises maintenance costs over time.
Refactoring	Refactoring is the process of improving the internal structure of code without changing what the software does from the user’s point of view.	Refactoring helps keep software maintainable over time, especially as products grow more complex.
Quality assurance (QA)	Quality assurance, usually shortened to QA, refers to the processes used to check whether software works as intended and meets required standards before release.	QA sits at the heart of product reliability and user trust, even if it is sometimes treated as a bottleneck rather than a discipline.

Internet, Advertising & Digital Platforms

The internet economy has developed a vocabulary around platforms, advertising infrastructure, data and digital distribution. These terms help explain how audiences are reached, how advertising is priced and how power is exercised online.

Key terms at a glance

Term	Definition	Why it matters
Walled garden	A walled garden is a closed digital ecosystem in which one company controls the platform, user access, data and often the commercial rules.	Walled gardens can create convenience and control, but they also raise concerns around competition, fees and access to customer data.
Platform risk	Platform risk describes the danger that a company’s growth or business model depends too heavily on another platform, such as Google, Apple, Amazon or Meta.	Businesses exposed to platform risk can be hit hard by algorithm changes, fee increases or policy shifts they do not control.
First-party data	First-party data is information a company collects directly from its own users or customers, rather than buying or sourcing it from third parties.	As privacy rules tighten and third-party tracking becomes harder, first-party data has become more strategically important.
Advertising technology (adtech)	Advertising technology, often shortened to adtech, is the software and infrastructure used to buy, sell, target, serve and measure digital advertising.	Adtech powers a large share of the internet economy, but it is also one of the sector’s most jargon-heavy and least transparent areas.
Programmatic advertising	Programmatic advertising is the automated buying and selling of digital advertising space, usually using software and real-time auctions.	It underpins much of the modern online advertising market, even if its mechanics are largely invisible to ordinary users.
Demand-side platform (DSP)	A demand-side platform, usually shortened to DSP, is software used by advertisers and agencies to buy digital ad inventory.	DSPs are a core part of programmatic advertising because they help buyers target audiences and manage campaigns at scale.
Supply-side platform (SSP)	A supply-side platform, usually shortened to SSP, is software used by publishers to sell digital ad inventory.	SSPs help publishers manage demand and maximise the value of their advertising space.
Cost per mille (CPM)	Cost per mille, usually shortened to CPM, refers to the cost of serving one thousand ad impressions.	CPM is one of the standard pricing models in digital advertising and is especially common in campaigns where reach matters more than immediate action.
Cost per click (CPC)	Cost per click, usually shortened to CPC, is a digital advertising model in which an advertiser pays each time a user clicks on an ad.	CPC is a common way of buying search and performance advertising, especially when advertisers want to pay for traffic rather than impressions alone.
Cost per acquisition (CPA)	Cost per acquisition, usually shortened to CPA, is a pricing or performance metric that measures how much it costs to acquire a customer or prompt a specific action, such as a purchase or sign-up.	CPA is widely used in performance marketing because it ties advertising spend directly to a commercial outcome.
Data clean room	A data clean room is a controlled environment in which different parties can compare or analyse data sets without directly exposing raw personal data to one another.	Data clean rooms have become more important as privacy rules tighten and companies look for new ways to measure audiences and campaign performance.

Games Industry Terms

Games have their own distinct commercial and operational vocabulary. These terms describe the business models, monetization mechanics and user metrics that shape how modern games are built and run.

Key terms at a glance

Term	Definition	Why it matters
Live-service	Live-service describes a game designed to evolve after launch through updates, seasonal content, events and ongoing player engagement.	Live-service models aim to extend the commercial life of a game and create recurring spending over time.
Live operations (live ops)	Live operations, usually shortened to live ops, refers to the ongoing management of a game after launch through events, updates, promotions, balance changes and community activity.	In many modern games, the commercial lifecycle depends less on launch day and more on how effectively a title is operated over time.
Free-to-play (F2P)	Free-to-play, usually shortened to F2P, describes games that are free to download or access, with revenue generated later through in-game purchases, advertising or other forms of monetization.	Free-to-play reshaped the economics of gaming by lowering the barrier to entry and shifting monetization deeper into the product experience.
Battle pass	A battle pass is a progression-based monetization system in which players unlock rewards over a fixed season, usually through gameplay or paid access.	Battle passes have become a central way of structuring engagement and recurring spending in live-service games.
Gacha	Gacha refers to a monetization mechanic in which players spend currency for a random chance of receiving characters, items or upgrades.	The model has been commercially powerful, especially in mobile games, but remains controversial because of its similarity to gambling-like reward structures.
Whale	A whale is a small subset of players who account for a disproportionately large share of spending in a game.	The term is blunt, but it reflects an important commercial reality in many digital games, especially where revenue is concentrated among a narrow group of high spenders.
Downloadable content (DLC)	Downloadable content, usually shortened to DLC, refers to additional game content released after launch, such as expansions, cosmetics, characters or missions.	DLC is one of the main ways games extend both their lifespan and their revenue base after release.
User-generated content (UGC)	User-generated content, usually shortened to UGC, refers to game content created by players rather than the original developer, such as mods, maps, skins or community-built experiences.	UGC can deepen engagement, extend a game’s life and create new platform-style dynamics inside gaming ecosystems.
Daily active users (DAU)	Daily active users, usually shortened to DAU, measures how many people engage with a game on a given day.	DAU is a basic measure of scale and regular usage in live and mobile games.
Monthly active users (MAU)	Monthly active users, usually shortened to MAU, measures how many people engage with a game over the course of a month.	Compared with DAU, MAU helps show the size of the broader audience and how habit-forming a product may be.
User acquisition (UA)	User acquisition, usually shortened to UA, refers to the process of attracting new players to a game, often through paid advertising.	Gaming companies depend heavily on their ability to acquire users efficiently and profitably.
Cost per install (CPI)	Cost per install, usually shortened to CPI, measures how much it costs to drive one app install through paid marketing.	CPI is a basic input in judging whether a mobile game’s user acquisition strategy is sustainable.
Average revenue per daily active user (ARPDAU)	Average revenue per daily active user, usually shortened to ARPDAU, measures how much revenue a game earns, on average, from each daily active user.	It is a key gaming metric because it links engagement directly to monetization performance.
In-app purchases (IAP)	In-app purchases, usually shortened to IAP, are items or features users buy inside a game, such as virtual currency, cosmetics or content packs.	IAP is one of the main ways free-to-play games generate revenue.
Rewarded ads	Rewarded ads are advertisements that players choose to watch in exchange for an in-game reward, such as currency or another chance to continue playing.	Rewarded ads offer a way to monetize non-paying users without relying entirely on more disruptive ad formats.
Soft launch	A soft launch is the limited release of a game in selected markets before a wider rollout.	Soft launches allow developers to test retention, monetization and performance before committing to a full release.
Hypercasual	Hypercasual games are simple, easy-to-learn games designed for broad appeal, short sessions and rapid onboarding.	The category became associated with cheap production, aggressive testing and advertising-led monetization, even if its prominence has shifted over time.

Fintech & Payments

Fintech sits at the intersection of software, regulation and financial infrastructure. Many of its key terms describe the hidden systems behind payments, banking and compliance rather than just the customer-facing products people see.

Key terms at a glance

Term	Definition	Why it matters
Embedded finance	Embedded finance refers to financial services built into non-financial products or platforms, such as payments, lending or insurance offered at the point of use.	It is blurring the line between technology companies and traditional financial institutions.
Open banking	Open banking is a framework that allows consumers to share banking data securely with third-party providers, usually through application programming interfaces.	It has enabled new products in payments, budgeting, lending and account aggregation.
Banking-as-a-service (BaaS)	Banking-as-a-service, usually shortened to BaaS, allows non-bank companies to offer financial products by plugging into regulated banking infrastructure provided by third parties.	It has become a foundational layer for many fintech business models.
Neobank	A neobank is a digital-first banking business that typically operates without a traditional branch network.	Neobanks have become one of the most visible consumer-facing expressions of fintech, although their economics vary widely.
Buy now, pay later (BNPL)	Buy now, pay later, usually shortened to BNPL, is a form of short-term credit that lets consumers split purchases into instalments, typically at checkout.	BNPL has grown rapidly by embedding credit into ecommerce, although it has also drawn regulatory scrutiny.
Payment rail	A payment rail is the underlying system or network used to move money between parties.	Payment rails are the infrastructure layer beneath many consumer-facing payment products.
Issuer processor	An issuer processor is the technology provider that helps a card issuer manage authorisation, settlement and other operational functions for card transactions.	Issuer processors are part of the hidden infrastructure that allows fintech companies to launch card-based products.
Acquiring bank	An acquiring bank is the financial institution that processes card payments on behalf of a merchant.	Acquiring banks sit on the merchant side of the payments system and are central to how card transactions are accepted and settled.
Interchange	Interchange is the fee paid between banks when a card transaction is processed, usually flowing from the merchant’s bank to the cardholder’s bank.	It is a crucial but often poorly understood part of the economics of card payments and card-linked fintech products.
Know your customer (KYC)	Know your customer, usually shortened to KYC, refers to the checks financial institutions use to verify customer identity and assess risk.	KYC is central to compliance, fraud prevention and anti-money laundering controls.
Anti-money laundering (AML)	Anti-money laundering, usually shortened to AML, refers to the rules, checks and systems designed to detect and prevent financial crime.	AML requirements shape how fintech products are built, monitored and regulated.
Fraud stack	Fraud stack is an informal term for the set of tools, models and operational processes a company uses to detect, prevent and respond to fraud.	In fintech, fraud management is not a side issue. It is a core part of the product and cost structure.

Artificial Intelligence & Machine Learning

Artificial intelligence has generated a fast-growing body of technical and commercial language. Some of these terms describe core model concepts, while others refer to how AI systems are adapted, deployed and maintained in practice.

Key terms at a glance

Term	Definition	Why it matters
Artificial intelligence (AI)	Artificial intelligence, usually shortened to AI, refers to computer systems designed to perform tasks that typically require human-like reasoning, pattern recognition or decision-making.	AI has become a catch-all term for a wide range of technologies, making clarity especially important.
Large language model (LLM)	Large language model, usually shortened to LLM, refers to an artificial intelligence model trained on vast amounts of text in order to generate and interpret language.	LLMs sit behind many of the generative AI tools now being used in search, productivity software and customer service.
Multimodal	A multimodal artificial intelligence system can work across different types of input or output, such as text, image, audio or video.	Multimodality is expanding what artificial intelligence tools can do beyond text alone.
Context window	A context window is the amount of information an artificial intelligence model can take into account at one time when generating a response.	Larger context windows can make models more useful for longer documents, more complex prompts and multi-step tasks.
Prompt engineering	Prompt engineering refers to the practice of designing prompts in ways that improve the quality, relevance or reliability of an artificial intelligence system’s outputs.	Prompt design can have a major effect on performance, especially when models are being used for repeatable business tasks.
Fine-tuning	Fine-tuning is the process of taking a pre-trained artificial intelligence model and training it further on a narrower dataset for a more specific task or domain.	It can make models more useful for specialist applications, although it also adds cost and complexity.
Retrieval-augmented generation (RAG)	Retrieval-augmented generation, usually shortened to RAG, is an artificial intelligence approach in which a model retrieves relevant information from external sources before generating a response.	RAG is often used to make artificial intelligence systems more useful in practical settings by grounding outputs in specific documents or datasets, rather than relying only on what the model learned during training.
Inference	Inference is the stage at which a trained artificial intelligence model is actually used to generate outputs or make predictions.	Inference is where artificial intelligence systems deliver value in practice, and where many of the real computing costs arise.
Agent	In artificial intelligence, an agent generally refers to a system that can carry out tasks in a more autonomous way, often by combining reasoning, software tools and multiple steps of action.	Artificial intelligence agents are one of the most heavily discussed ideas in the current market, although the term is often used loosely and inconsistently.
Hallucination	A hallucination occurs when an artificial intelligence model produces information that sounds convincing but is inaccurate, fabricated or misleading.	Hallucinations are one of the main reasons artificial intelligence outputs still need checking, especially in high-stakes settings.
Vector database	A vector database is a database designed to store and retrieve numerical representations of data, often used to support similarity search in artificial intelligence applications.	Vector databases are commonly used in systems that need to find relevant information quickly, including many retrieval-augmented generation setups.
Model drift	Model drift describes the decline in an artificial intelligence system’s performance over time as real-world conditions or data patterns change.	Drift is one of the reasons artificial intelligence systems need monitoring and maintenance after deployment.
Synthetic data	Synthetic data is artificially generated data created to resemble real-world data.	Synthetic data can help with training, testing and privacy-sensitive applications, although its usefulness depends on how realistic and representative it is.

The language of the tech economy will keep changing. But while the jargon evolves, the need for clarity does not. The more these industries shape business, media and everyday life, the more useful it becomes to understand the terms that structure them.

Peter Franks

Founder & Editor, No Latency

Peter writes long-form analysis on technology, gaming and artificial intelligence – focusing on the systems, incentives and strategic decisions shaping the modern software economy.

He has spent 20+ years working with software and games companies across Europe, advising founders, executives and investors on leadership and organisational design. He is also the founder of Neon River, a specialist executive search firm.

Connect on LinkedIn

Built by No Latency