Technology Glossary

Glossary of terms

Agentic AI

An AI system that plans and carries out multi-step tasks by calling tools, rather than only producing text. In enterprise deployments an agentic system operates within explicitly defined tool permissions, escalates when confidence is low, and logs every action it takes.

Prompt injection

An attack in which untrusted content (a document, web page or email that an AI system reads) contains text the system treats as an instruction. Because the system acts with its own authenticated permissions, a successful injection can lead to data disclosure or unintended actions.

Retrieval-augmented generation

A technique in which an AI system retrieves relevant documents from an approved source before generating an answer, and cites them. It grounds responses in verifiable material rather than relying on the model's internal knowledge.

Offshore Development Centre

A dedicated engineering team operating from a supplier's location but working exclusively on one client's roadmap, under that client's technical direction, with its own delivery manager and governance.

Strangler pattern

An incremental modernisation approach in which new functionality is built behind a stable interface while the legacy system continues to serve everything not yet replaced, allowing the old system to be retired gradually rather than in a single cutover.

Infrastructure as code

Defining servers, networks and cloud resources in version-controlled configuration files rather than through manual console changes, so environments are reproducible, reviewable and disposable.

Recovery Time Objective

The maximum acceptable period between a disruption and the restoration of a service. It is set by business tolerance and should be proven through rehearsal rather than estimated.

Recovery Point Objective

The maximum acceptable amount of data loss, expressed as a period of time. An RPO of one hour means the organisation accepts losing up to one hour of data in a recovery scenario.

Intelligent process automation

The combination of rule-based workflow automation with AI applied to the steps requiring reading, classification or judgement, for example extracting data from an invoice before routing it for approval.

Digital twin

A virtual model of a physical asset or process, fed by live operational data, used to monitor current state and simulate the effect of proposed changes before committing to them physically.

Idempotency

A property of an operation that produces the same result whether performed once or several times. It is essential in integrations, where network failures cause messages to be delivered more than once.

Zero trust

A security model in which no user, device or network location is trusted by default. Every request is authenticated and authorised, and permissions are granted at the minimum level required for the task.