A server whose function is receiving code updates to a repository and distributing builds.
"We organize all of our commits in a code repository manager that helps track progress and debug problems caused by forked code."
Submission of software code to a version-controlled repository, usually grouped around a specific problem which has been solved (such as a bug fix).
"The team committed several changes to the app today, focused on patching security hole and improving pairing with bluetooth devices."
A problem where the problem is understood, but not how to solve it.
"Anticipating business user behavior is a complex problem—there are a lot of interconnected factors that we can't fully understand because we can't fully see the enterprises our customers exist within."
A mindset that allows machines and humans to work together to solve real-world problems, using Decomposition, Abstraction, Patterns, Algorithms, and Programs.
"Using computational thinking, we can break down the task of pouring a glass of water into a series of smaller, machine-friendly tasks, such as 'determine the position of the glass relative to the jug'."
The computer chips and processing power needed to build and run AI systems. More compute generally means more powerful AI.
As AI models get more complex, the amount of compute needed to build and run them keeps growing.
The processing power provided by computer chips, used to run software, train AI models, and power digital services. More compute generally means faster, more capable systems.
As AI models get more complex, the amount of compute needed to build and run them keeps growing.
Lightweight packages of application code along with just the things the application specifically depends on, such as specific versions of programming languages and libraries required to run a component of software being programmed.
"By using containers, the developer was able to ensure the software ran in any computing environment across the company, simplifying the deployment process."
The amount of information a model can “hold in mind” at one time.
"Because the context window was limited, the chatbot forgot details from earlier in the conversation."
The practice of developing software in short, repeated release cycles with the intent to consistently raise its quality, performance or utility—rather than wait for the next major edition of the software before making fixes or improvements. (This is sometimes used interchangeably or in conjunction with ‘continuous improvement.’)
“We always want our customers to have the best, most current version of our app possible, so instead of doing a quarterly update, we practice continuous delivery.”
Continuous Delivery is the practice of developing software in short release cycles where the software is incrementally updated and manually deployed; as opposed to continuous deployment, wherein the software is automatically deployed
"We always want our customers to have the best, most current version of our app possible, so instead of doing a quarterly update, we practice continuous deployment."
The practice of regularly collecting both structured and unstructured assessments and critiques to apply to future development.
"Since we implement continuous feedback, our developers feel like they know our users better."
Computer server(s) and program(s) which automatically build and update software according to predefined rules.
"By implementing a CI server in our development pipeline, we automated integrating code changes, which increased the speed and efficiency of our software development process and lowered stress."
The usage of automated and manual assessments of a program’s efficiency and stability as soon as the code changes
"Continuous testing is your only safeguard against software failure."
AIs designed to create human-like interaction via chat or voice dialog using natural language processing and language learning models like GPT.
"The conversational AI field has reached an inflection point as the size and accuracy of LLMs like GPT have become sufficient (and sufficiently accessible) to allow everyday productivity improvements."
The use of chatbots, messaging apps, or voice assistants to facilitate shopping and customer interactions.
"Customers ordered directly through WhatsApp thanks to conversational commerce tools."
The idea of being able to have a conversation 'with your data,' using a chat-based interface for data science tools.
"We implemented a conversational data strategy for our executives and managers, so that they don't have to put in a request for data scientist for simple questions about our performance."
A branded term (popularized by Microsoft) for an AI assistant that works alongside humans, helping draft text, analyze data, or perform tasks within familiar software.
"We used Copilot in Word to draft a proposal outline, then edited it ourselves for accuracy."
The price of acquiring and using a dataset—covers per-unit cost, pricing model (one-time, subscription), and usage limits. Distinct from research cost (gathering original data) and analysis cost (extracting value).
The dataset's cost was high, but spread across the team's projects, it paid for itself in a quarter.
A breakdown of all costs involved in running an organization.
"The cost structure of the business included fixed costs like facilities rent, but also variable costs such as cloud storage and hosting fees."
AI systems that solve problems by breaking them down into logical, sequential steps, making their decision process more transparent and reliable.
"The reasoning agent analyzed the legal document and highlighted clauses that needed extra review, then went back after reading the whole document and made suggestions for each clause.
The geographic area a dataset describes—global, regional, or local. A USA-only dataset has narrower coverage than a global one.
The coverage was perfect for U.S. operations, but useless for the Asia expansion.
The ways a business connects with its customers.
"Call centers, help desks, and even community forums are all customer channels being changed by generative AI."
Systems that collect, unify, and organize customer data from multiple sources to build a single customer profile.
"Our CDP pulled together data from email, web, and sales so marketing could personalize campaigns."
Ways a customer's 'jobs to be done' could be easier, faster, or otherwise improved, as described in a Value Proposition.
"Our value prop focuses on a key customer gain: faster time to market for enterprise app developers."
Emotional, social, or functional tasks a user may need to complete, as addressed in a Value Proposition
"Our customer's 'jobs to be done,' like 'create a blog post,' 'come up with social media hashtags' and 'create a press release' are directly mapped to buttons in the app."
Any risks and obstacles a customer may wish to avoid or minimize in the course of their 'jobs to be done', as described in a Value Proposition.
"We relieve the customer's pains for international payments—we make tax, compliance and currency conversion seamless so they can just focus on their products."
Fictional profiles of users, customers, or other stakeholders who exemplify the kind of individuals an organization wants to serve.
"One of our customer personas is Saul, a 35-year-old single father who works as an attorney and lives in the city center."
The nature of the relationships created between business and its customers, as described in a Business Model.
"Do we have to create all value in our relationships with customers like a traditional widget company, or could we co-create with them like Starbucks does when they crowdsource flavor ideas?"
Dividing a company's customers into groups that reflect similarity among customers in each group, with an eye to relate to each segment in ways that maximize the value of each customer to the business.
"As our user advocate, I identified three customer segments: convenience-seeking young professionals, value-conscious families, and tech enthusiasts. This allowed us to tailor our features and plans to each group."
Short for cybersecurity—the practice of protecting computers, networks, and data from attack, damage, or unauthorized access. Also used as an adjective to describe digital threats or capabilities.
Organizations invest heavily in cyber defenses to protect sensitive data from hackers and other threats.
The merging of organic and inorganic, or human and machine, to create a third type of entity. Cyborgs feature in science fiction (like Robocop). The cyborg concept is also a mental model of humans' relationship to technology as co-constitutive, wherein humans don't just make tech, but it remakes and reshapes them. Implants, exoskeletons, augmented reality, and even lower-tech connections like humans aided by smartphones are all cyborg relationships.
"Humans and their vehicles can be thought of as cyborgs, where the human is both extended by their interface with the machine, but also changed by it—we started to imagine new ways to navigate, more like 'robo-shoes' and less like 'cars.'"
A person who uses the mental model of a human-machine 'cyborg' to study the impact of technology on individuals, cultures, and the planet. Cyborg Anthropologists help businesses and organizations understand the impact of complex issues like artificial intelligence, data ethics, the future of work, and virtual worlds.
"A health company hired a cyborg anthropologist to help them understand how 'augmentation' implants like optical-nerve-to-computer interfaces could positively or negatively affect users' confidence in themselves."
A short, informal meeting at the beginning of the day to share current individual project statuses and make requests for support.
"Fortunately, during the daily standup, I found out the feature that I was planning on rushing out the door is no longer as urgent, so I can refine it a bit more."
DALL-E is an AI-powered image generator developed by OpenAI. It is a variant of the GPT-3 language model, trained on a diverse dataset of text and images to generate original, high-resolution images from textual descriptions. DALL-E can create a wide range of images, from photorealistic to highly imaginative, based on input text that describes the desired image. (ChatGPT wrote this definition of DALL-E.)
“We were imagining new characters for our video game, and we used DALL-E to quickly turn the character bios into sketches.”
A user interface or feature that tricks users into doing things that may not be good for them, often for the benefit of the developer. This term is also sometimes used to describe ongoing problematic patterns in business or culture.
"The biggest tax software maker is known for its dark patterns, such as tricking users into selecting a paid plan when free plans were available, by making it hard to find the free option."
A part of the internet that is not indexed by mainstream search engines and which usually must be accessed through special software (like Tor), known for attracting illegal content, communication, and transactions.
"Hackers leaked the passwords onto the dark web to attract purchasers who intended to use them for fraud and identity theft."
Software tools which show key performance indicators related to business or user needs.
"During meetings, the software team leader would often show a project management dashboard that displayed the current backlog of programming tasks and the team's velocity in finishing them."
Raw facts, numbers, or observations that can be processed or analyzed by computers.
"The survey produced data on customer preferences that we used to shape product design."
Examining and transforming data into information—finding meaning and/or insights through scrutiny.
"Through rigorous data analysis, ProPublica found that the 'predictive sentencing systems' used by judges across the country were regularly over-sentencing black people who turned out to be lower risks than their white counterparts."
The models, policies, rules, and standards that govern which data is collected and how it is stored, arranged, integrated, and put to use in data systems and in organizations.
"When we're merging with AcmeCo, we'll need to update data architectures in both companies to revisit decisions about how we store and process data, and make sure that standards, policies and expectation-setting for users are aligned."
Data that is not actively moving from one device or system to another; for example, files saved to a local hard drive.
"Our company's mindset is very 'Data at Rest:' files in a filing cabinet, rows in a spreadsheet, events on the wall calendar: data is static, only available in one place, and stays the same until a human modifies it."
A measurable trait of a dataset that helps compare it to other datasets—like relevance, quality, freshness, or cost. Used to evaluate whether data is fit for a specific purpose.
She built a rubric of data attributes to score each dataset before committing to a subscription.
An incident where sensitive or confidential information is accessed or disclosed without authorization, either by mistakes, negligence, or malicious actors (hacking).
"In 2017, a credit reporting agency experienced a data breach that exposed the personal information of over 147 million customers, including Social Security numbers and birth dates."
Comprehensive, organized lists of what datasets are available from an organization or other source.
"This data catalog indicates all of types of customer and user data available within the company, such as website actions, purchases, and contact information."
A large facility filled with servers and networking equipment used to store, process, and distribute data and digital services.
When we stream a video or use a cloud app, the content is being served from a data center somewhere in the world, using their local electricity, labor and water (for cooling).
A set of signed, legally-binding agreements that define the terms and conditions of interactions between two or more parties. Data contracts may specify the roles and responsibilities of each party, ownership of data and usage rights, acceptable levels of privacy, provenance, data sovereignty, methods of data validation, and storage/disposal requirements.
"When we partnered with another company to process our user's inputs into our chatbot, we had to set up data contracts to make sure our users' privacy preferences were honored and that their data stayed in the right legal jurisdictions."
A strategy that uses customer and performance data to guide marketing decisions.
"Data-driven marketing helped us increase conversions by targeting specific types of customers more precisely."
The mindsets, frameworks, values, rules and principles for responsibly gather, manipulating, applying and monetizing data from users and organizations.
"A lot of AI developers are exploring ideas which seem good—like providing increased access to medical information. However, they don't always have a strong data ethics strategy to protect against leaks, give consistent analysis or disclose an AI system's limits to unsuspecting users."
Something recorded or logged from a machine systgem without a specific intent in mind.
"Our data exhaust includes keycard logins, the individual devices that connect to the office wifi, and security camera feeds."
The process of checking, correcting, and standardizing data for accuracy and consistency, including verifying formats, creating unique identifiers, and categorizing information.
"We have a lot of responses from past marketing efforts, but the data hygiene is very poor—none of the questions asked of users were the same, and the contact information is not checked for accuracy or typos. It would be cheaper to start over than fix it."
The practice of checking, correcting, labeling, and normalizing data. Common activities related to data hygiene include checking for accuracy, ensuring formats are the same in each dataset (such as the format for date & time), determining or creating a unique identifier (such as an email address or phone number, which allows combining one dataset with others), and categorizing data.
"We have a lot of responses from past marketing efforts, but the data hygiene is very poor—none of the questions asked of users were the same, and the contact information is not checked for accuracy or typos. It would be cheaper to start over than fix it."
Data in Motion is dynamic, 'live', moving and changing in real time, as opposed to a static file saved on one machine.
"Data in motion can be thought of like water—a ‘stream’ like a live video feed or a ‘flow’ of stock market data."
Storing large amounts of data to be processed later; often refers to data 'at rest.'
"Facebook's Social Graph is an enormous data lake that stores countless pieces of data on every user for various analyses, some of which were perhaps not obvious at the time of collection."
Spaces designed for parties to buy, sell, and lease data to each other, including both broad and highly-focused datasets.
"The Amazon Web Services Marketplace offers a very large range of datasets for subscription, with both data providers and data processors; many companies list AI-optimized training data in the marketplace."
Streams of data from various sources aggregated into one flow. Related to the concepts of data lakes and data in motion.
"Through aggregation of many different data streams, we now have a 'data river' of nearly all customer transactions and interactions across all our platforms, from Amazon to Zoom."
The concept that data should be subject to the laws of the country in which is it collected, stored, and processed while it is within that country.
"When we expand into European countries, we have to make sure we respect the data sovereignty requirements which necessitate storing data about EU users in a member state rather than our US servers."
People or machines who disclose data
"The 'historical AI video' based on data subjects like Steve Jobs couldn't consent to their data being used for generative AI training, because generative AI was not something people understood at the time they were recorded."
Classification systems used to group and sort data; usually formal and applied by organizing bodies.
"The Dewey Decimal System is one of the most well-known data taxonomies of the 20th century, used to organize topics in libraries and research."
Breaking down a complex problem into several simpler problems.
"When we decompose hailing a rideshare service, we see that it can be broken down into simpler problems, including determining journey start and end points, finding an efficient route between the two, and calculating an appropriate fare."
Believable impersonation of a human's likeness using advanced machine learning/generative AI tools, usually without consent, to spread misinformation, engage in parody or synthesize content (such as pornography or music). Deepfakes usually refer to photos, audio, or video and may be coupled with text written a particular style and/or distributed on hacked or impersonated social media.
"BuzzFeed shared a video of former U.S. President Barack Obama vulgarly insulting Trump—only to go on to reveal that the video was a deepfake and to caution people about 'truth' on the internet."
An AI model developed by a Chinese company of the same name. It gained attention for performing competitively with leading Western AI models, often at lower cost.
DeepSeek was widely discussed as an example of how quickly AI development is advancing outside of the United States.
How much a dataset has been organized, defined, and categorized. Structured data (rows and columns) is easy to query; unstructured data (raw text, video, images) needs more work. Estimates suggest only about 5% of available data is structured.
The dataset was rich but had a low degree of structure—weeks of cleanup before it was usable.
The variety and number of distinct data points per record. A customer record showing only login times has low depth; one showing every action on the site has high depth.
The depth let the team see not just that users dropped off, but exactly where in the flow.
A human-centered approach to innovation that focuses on understanding the problems and goals of potential customers or users as a method of defining and testing new product ideas.
"We used design thinking methods, such as user observation and empathy experiences, to ensure that we solve real problems for real people, rather than just spending our time making products we thought might be good but weren't necessarily helpful."
A software development cycle that allows companies to produce, release, and monitor iterative software updates on vastly reduced timetables, compared to traditional development cycles.
"Our DevOps team constantly updates and deploys software as new features, fixes and use cases develop—which was a big mindset shift from older, annual releases which were considered 'done' once launched."
Development Security Operations or DevSecOps is the practice of including security and compliance departments from the beginning of development processes, rather than as a sign-off stage after most architecture, feature and data choices have already been made.
"Before we implemented DevSecOps, we launched fast and often—only to have to postpone or roll back releases once our security and compliance teams saw problems. Now we include security stakeholders at every stage of production."
A set of people and capabilities which allow for the ongoing creation of digital products, services and features, usually with a 'continuous improvement' mindset and clear development operations (DevOps).
"You can tell AcmeCo is serious about becoming more of a software company instead of just outsourcing tech work—they have a digital factory to help scale teams' good ideas to something reliable and consistent."
True understanding of the thinking, data, business models, tools and skills needed for people and companies to digitally transform, ’go digital,’ or offer digital products.
"We're upping our teams' Digital Fluency with a new mentoring and education program."
Digital Twin: a machine data model of a physical object, such as a jet engine, a person's body, or a building. This model is built using precise measurements and is usually 'brought to life' with live sensor readings and predictive machine learning algorithms intended to predict aging, performance, failure risks, and other information.
"The car has a digital twin of the motor and uses an algorithm to adjust recommended maintenance intervals based on how much stress the motor is experiencing."
There is no rigid, 'official' definition, but Digital Value propositions use exponential technologies to offer speed, scale, or ease which is not possible with a conventional value proposition such as one focused on physical assets or human services.
"Uber's original digital value proposition was to offer you 'your own private driver'. Offered to everyday middle-class users in San Francisco, the digital platform marketplace drastically lowered the cost of a private luxury ride."
Analog: 1) the use of signals or information represented by a variable physical quantity, like position, voltage, or wave frequency; 2) a contrast or predecessor to the use of digital computer technologies.
Digital: 1) the use of numbers and mathematical representations (usually the binary 1 or 0) to capture and convey information; 2) software-defined value, offerings, and organizations.
“We used to manage all of our customer relationships through team members in our branches, but now we also work with our customers digitally—via APIs, chat systems and on-demand support documentation.”
Inclusion of unique data points (such as pixels, special characters, or key phrases) in AI-generated or -edited content. Digital watermarks can be used to spot plagiarism, manage intellectual property, or lessen the risk of 'deepfakes' and other problematic uses of AI.
"OpenAI announced work on a digital watermark to ensure student use of their tool was properly cited, addressing a key concern of academia and educators."
A mode of thinking which generates ideas that are not necessarily aligned with a group, organization, or the status quo. This is a desirable kind of thinking in ideation and early stages of innovation.
"Our company struggles with divergent thinking. Although we're quick to say we want new ideas, we're even quicker to say why they won't work. It's why we give the innovation team non-revenue metrics so that good ideas don't get written off before they've been matured a bit."
Configuring and training AI systems for use in particular contexts. Examples include training writing systems for a specific company's tone and vocabulary, ensuring medical systems respond with culturally-appropriate answers, or configuring legal tools for particular jurisdictions.
"AcmeCo trained their large language model in the specific terms used by their users and customers to make their customer service tool more effective and relevant."
Technology that has both everyday positive applications could also have harmful uses. In Cyborg Anthropology, this is summed up by the phrase "technology has no allegiance to its origin story."
AI is a dual-use technology: the same system that helps a doctor diagnose disease could also be used to plan a cyberattack.
A special kind of non-fungible token with dynamic contents which can be updated under certain conditions.
"An artist released a dNFT that layers unique, algorithmically-generated art which displays different results based on the current weather conditions of the user (like an advanced camera filter)."
Artificial intelligence systems which run partially or fully near the sensors or end devices users interact with rather than solely on central servers. Edge AI examples include autonomous vehicle decision-making systems, smartphone facial recognition software, and security camera motion-sensing systems.
"Some of Apple's image recognition occurs on users' devices so that their photo data is not passed through cloud servers and cannot be accessed by Apple or hackers who might breach a cloud system."
Numerical representations of text or other data that make it easy to compare similarity.
"We used embeddings to quickly find documents related to the customer’s support ticket."
A contract outlining the terms and conditions that a user must agree to in order to use a specific app or digital service.
"I know I probably should read all the fine print in a EULA, but I usually just click 'agree' so I can move on to using the app. But now I'm wondering if I gave away rights for my content to train an AI system."
Collections of User Stories, usually grouped according to theme as part of a feedback loop in Agile development and Devops.
"The team grouped all the user stories relating to how users searched for data into an epic."
Expert systems attempt to mirror a human expert's knowledge and decision-making power through absolute knowledge recall from a database. These systems can be thought of as flow charts or 'if this, then that' decision trees.
"Generative AI systems should sometimes be paired with more traditional 'expert systems' to ensure their answers on critical topics like medicine are grounded in facts—in the way a human doctor uses a diagnostic manual to carefully rule out possible diseases and aid in common diagnoses."
Artificial intelligence (AI) or machine learning (ML) models that can be sufficiently reverse-engineered to illustrate why a particular response to a query was given. In lay terms, a human can often (but not always) 'explain their thinking' when questioned about a decision, but not all AI/ML models can due to either their proprietary nature and/or the mathematical processes within.
"Conversational AI like ChatGPT is not always explainable AI, making it unsuitable for situations where decision-making must be explained, such as credit scoring, hiring decisions or academia."
A fundamental change in the way a business creates and delivers value, using network effects that (usually) come with new, digital, technologies.
"Uber's ridesharing business model was an exponential innovation because it harnessed network effects: as more drivers and riders joined the system, the value of the network grew exponentially rather than linearly."
Government rules that restrict which products or technologies can be sold to other countries.
Export controls have slowed China's AI development by limiting its access to the best chips made by American companies.
Extended Reality or XR is an umbrella term for any technology that blends the digital world with the analog physical world.
"Our extended reality team is working to make sure our products are usable in virtual worlds, whether VR, augmented reality, the metaverse or video games."
Concern that it won't be possible to keep up with with technology-related job requirements, economic pressures and/or social changes.
"Today I saw a presentation on how all future hires will have to have coding and AI experience—I'm having major FOBO. Does this mean I'm not relevant anymore?
Anxiety about not being able to participate in something exciting or new, or being left out or left behind.
"I see all these AI startups getting tons of cash and attention and talent, and I feel a lot of FOMO—did we time the market wrong? Or are we just getting started?"
Retraining a generative AI model’s weights with specialized examples until it “speaks” like the right kind of organization or answers questions focused on a particular domain.
"The model was pretty good, but it kept returning examples in US English, so we fine-tuned it to use UK English instead."
Crowd-based, informal classification of data that occurs 'organically' and emerges from users, rather than being imposed from a central power like a taxonomy.
"Twitter's hashtags are the ultimate folksonomy—all hashtags are created or chosen by individual users and emerge as trends."
How often a dataset is updated. Live data refreshes constantly (tweets, sensor streams); periodic data refreshes on a schedule (daily report, annual filing). The right frequency depends on the use case.
Quarterly reports didn't cut it for high-frequency trading—they needed second-by-second frequency.
Something directly and intentionally measured, like a list of stock investments or an image file on a camera.
"The fundamental data about the company includes its quarterly reports, stock performance, and degree of volatility in the marketplace."
Artificial intelligence models that synthesize new data (like text, music, video or images) based on patterns from existing datasets. These models are designed to create new and unique content resembling the input data their algorithms were trained on, like ChatGPT or DALL-E.
"Generative AIs are increasingly being used to replace rote work at our company—but we haven't really addressed how that will affect entry-level jobs."
Software that uses AI to draft marketing copy, blog posts, or ad content.
"The team used a generative AI-powered copywriting tool to create ad variations quickly."
Gradient descent is a step-by-step method used in machine learning to improve a model by reducing its errors. It does this by adjusting the model's settings based on how much it is "off" from the desired outcome. The algorithm calculates the direction and amount of adjustment needed and gradually moves the model towards the best possible performance.
"We used gradient descent and user feedback to tune the model to find the answer most likely to please people using the chatbot."
A data structure made of nodes and connections (edges) that shows relationships, often used in knowledge graphs or social networks.
"The AI used a graph of customer interactions to recommend new products."
a model/strategy that combines a database retrieval mechanism with a language generation model to aid in 'grounded' responses (factual, relevant answers), and avoid hallucinations (false answers).
"In order to make sure answers relevant to users, based in the articles our newspaper had actually written, and accurate, we chose a Graph RAG model for our 'fact-checker bot.'"
The practice of ensuring that generative AI tools return results that are accurate ('grounded' in facts) rather than just those which are statistically probable or pleasing to a user.
"OpenAI's ChatGPT can now provide citations of its sources so that users can understand the context of its answers."
A control or safeguard placed around AI systems to limit harmful or undesired outputs.
"The platform added AI guardrails to block inappropriate language in customer chats."
Rules or filters that limit what an AI can say or do, usually for safety or compliance.
"Guardrails prevented the bot from giving financial advice outside approved guidelines."
In the context of AI, a hallucination refers to a situation where an artificial intelligence system generates or perceives information that is not based on real or accurate data. It is an erroneous or false perception or output produced by the AI system.
"ChatGPT thought that in addition to owning Causeit, Inc. that I am a screenwriter for Marvel movies. So I said, 'where're my royalty payments?'"
