Knowledge
Knowledge HubConcepts and strategic researchGlossaryDefinitions across six domainsFrameworksStructured strategy methodsTopicsConnected subject areasLibrary
PlaybooksPractical strategy guidesBooksBooks and long-form researchOperator VaultMember resources and toolkits
Explore Discover, the Knowledge Hub, playbooks, and the Operator Vault.
Glossary
AI agent
An AI agent is a software actor that pursues a goal by perceiving state, deciding what to do next, and acting — often over multiple steps — within an environment of tools, data, and policies. Many contemporary agents are built around large language models and tool calling, but an agent need not use an LLM, call external APIs, or take multi-step tool actions in every design. In TSS usage, “AI agent” names the technical unit; broader product suites and platforms are discussed under Agentic AI, and enterprise decision rights under the Agentic Operating Model.
Why it matters
Agents change the risk profile of AI from “answer quality” to “action quality.” When an agent can write tickets, move money, change configurations, or message customers, governance must cover permissions, audit, escalation, and evaluation of trajectories — not only prompt wording.
Industry language is not standardized. Some products labeled “agent” are single-turn assistants; some agentic platforms mix agents with deterministic workflows and conventional software. TSS keeps a practical distinction: AI agent = technical actor; Agentic AI = broader initiative-taking systems and product category; Agentic Operating Model = TSS governance frame. Treat that as an operational classification convention, not a claim that the industry shares one definition.
Example (enterprise): An IT agent receives “provision sandbox access for contractor X,” checks identity policy, opens a temporary role in the IAM system, posts a ticket comment, and stops for human approval if the role exceeds a privilege threshold. Failure modes include wrong tool choice, incomplete state, and acting outside policy — not merely a bad paragraph of text.
Mechanisms
- Goal and state — a task objective plus memory or workspace of what has happened so far.
- Decision loop — planning or reactive selection of next actions (model-based, rule-based, or hybrid).
- Action channel — tools, APIs, UI automation, or other effectors when the design requires environmental change; not every agent design uses external tools.
Boundaries
- AI agent ≠ chat completion alone (a single response without an actor loop).
- AI agent ≠ Agentic AI as exact synonyms; TSS uses agent for the unit and agentic for the broader class (convention, not universal law).
- AI agent ≠ Agentic Operating Model (AOM is TSS enterprise design language for decision rights and operating design).
- “Agent” alone is too broad as a glossary alias — reserve AI agent for this technical sense.
Related topics
None linked yet.
Related concepts & frameworks
Related Discover collections
Further research
No related resources yet.