Glossary
The vocabulary used across IOanyT Academy, defined once and used the same way everywhere.
- Agent washing
- Marketing an existing product, such as a chatbot, an assistant or robotic process automation, as "agentic AI" when it lacks substantial agentic capability. Gartner uses the term for this rebranding.
- AI washing
- Making false or unsubstantiated claims about how much, or how well, a product or company uses AI. US regulators have brought enforcement actions over such claims.
- Learn more
- Annex III (EU AI Act)
- The EU AI Act list of high-risk use cases, including credit scoring, employment, education and insurance pricing. Under the 2026 Digital Omnibus agreement, its high-risk obligations apply from 2 December 2027.
- Learn more
- Bounded extraction
- A pattern in which a language model fills a strict, predefined schema, code validates the result, and a person takes over when the output is unsure or invalid. Level 2 of the Graduated Exposure Ladder.
- Learn more
- Build time and run time
- Build time is when software is designed, written and tested. Run time is when it serves real users and makes real decisions. Compiled AI uses language models at build time and deterministic code at run time.
- Learn more
- Compiled AI
- Compiled AI is a way of building software in which large language models write, test and harden the logic at build time, and deterministic code runs it in production. The model does the engineering; reviewed, versioned code does the executing, so the same input produces the same output every time.
- Learn more
- Deterministic AI
- "Deterministic AI" describes AI-built systems whose production behaviour is deterministic: the same input always produces the same output, because the decision path runs as fixed, reviewed logic rather than a fresh model generation on every request.
- Learn more
- Deterministic execution
- Running a decision path as fixed, reviewed, versioned code, so that identical inputs always produce identical outputs and every result can be traced to the logic that produced it.
- Learn more
- Golden dataset
- A set of real examples with answers your own team has agreed are correct, used to measure a system's accuracy before go-live and after every change.
- Graduated Exposure Ladder
- IOanyT's framework of five levels for deciding how much language-model behaviour is allowed at run time: Level 0 build-time only, Level 1 internal assist, Level 2 bounded extraction, Level 3 constrained conversation, Level 4 promotion.
- Learn more
- Proof Sprint
- IOanyT's fixed-scope 3–4 week pilot that compiles one workflow into deterministic logic and reports its run-to-run consistency, accuracy against the client's labelled examples, and cost per case.
- Learn more
- Regression evaluation
- Re-running a fixed set of test cases after every change to a system, to catch behaviour that got worse before it reaches users.
- Run-to-run variance
- Differences between a system's outputs when exactly the same input is processed more than once. Common in language-model outputs; absent from deterministic code.
- Learn more
- Temperature
- A sampling setting that controls how much randomness a language model uses when choosing its next word. Temperature 0 means always choosing the most likely option, which reduces variation but does not guarantee identical outputs in production.
- Learn more