Glossary
The vocabulary of agent-to-agent commerce.
43 terms, drawn from Teleperson's Research Library. Browse by section, search the page, or follow each term back to the paper that uses it most.
Architecture & protocols
The technical substrate of agent-to-agent commerce — how agents discover each other, talk, transact, and prove what happened.
Agent-to-agent (A2A) commerce
An architecture where AI agents acting on behalf of consumers transact directly with AI agents acting on behalf of brands, mediated by a neutral marketplace layer.
Model Context Protocol (MCP)
Anthropic's open protocol (released November 2024) for connecting agents to external tools and data sources. The de facto standard for the agent-to-tool integration layer.
A2A protocol
Google's Agent2Agent protocol (April 2025) for inter-agent communication. Defines the wire format and basic vocabulary for agents to talk to each other across platforms.
Agentic Commerce Protocol
OpenAI's commerce-vocabulary protocol for agent-to-agent transactions. Competes with Klarna's Agentic Product Protocol and merchant-side standards.
KYA (Know Your Agent)
Identity standards for verifying an agent's authorisation to act for a specific principal. Analogous to KYC for humans. The under-settled layer of the agentic stack.
Topology inversion
The shift from one-to-one (each consumer talking to each brand) to one-to-many (one consumer agent transacting with many brand agents through a hub).
Bring-your-own-bot pattern
Marketplace design where any agent can participate via standardised protocols, rather than requiring participants to use the marketplace operator's specific agent.
Connective tissue
The neutral marketplace layer enabling agent-to-agent commerce; the open alternative to closed-platform agent ecosystems.
Trust, identity, and receipts
What makes an agent transaction accountable, contestable, and reusable as evidence — the trust layer Teleperson and the rest of the category are building.
Signed receipt
A cryptographically signed artifact produced at the end of every agent-to-agent transaction. Documents what was agreed, by whom, on what evidence — and is admissible in dispute resolution.
Negotiation trace
The structured record of an agent-to-agent negotiation: stated objective, counter-offers, supporting context, agreed terms, verified outcome. Categorically richer than a clickstream.
Audit trail
The full record of an agent's actions, accessible to its principal for review and dispute. The substrate for both visitor trust and regulatory accountability.
Trust layer
The infrastructure (identity, receipts, reputation, watchers) that makes agent transactions accountable. Distinguishes a deployable agent from a demoware one.
Watcher
A monitoring agent that supervises another agent's actions, classifying high-stakes operations and enforcing principal-confirmation requirements before they bind.
Bounded authority
The scope of action an agent is authorised to take on behalf of its principal. Narrow scopes are safer; broad scopes shift risk to the principal.
Confirmation behaviour
The pattern of an agent checking back with its principal before executing actions that exceed certain thresholds. The operational mechanism that makes scope errors catchable.
Single-step authority
A pattern where the agent executes one bounded action per principal authorisation, rather than chaining multiple commitments without re-confirmation.
Roles & parties
Who's who in an agent-to-agent transaction. Most current liability discussions collapse this into two parties; there are four.
Principal
The human or institution on whose behalf an agent acts. Grants authority, defines scope, and benefits from (or is bound by) the agent's actions.
Consumer agent
An AI agent acting on behalf of an individual. Atlas, Teleperson's AI assistant for Mac, is a consumer agent.
Brand agent
An AI agent acting on behalf of a company. Customer-service-agent vendors and brand-side platforms ship these.
Agent provider
The company that builds and operates an agent. Responsible for the agent's behaviour staying within its declared capabilities.
Marketplace operator
The neutral layer where two agents transact. Defines the protocol, accepts identity claims, mediates disputes, and produces signed receipts. Teleperson is building toward this position.
Counterparty
The party on the other side of an agent-to-agent transaction — typically the brand whose agent accepted the consumer's offer (or vice versa).
Design partner
An early customer co-defining an agentic deployment. Teleperson runs a small design-partner cohort to ship the first agent-to-agent integrations.
Liability, regulation, and policy
The legal and regulatory questions that autonomous, multi-agent systems raise — and that conventional AI law does not yet answer.
Liability allocation
The framework determining which party (principal, agent provider, marketplace operator, counterparty) bears loss in a disputed agent transaction. No settled answer exists in current law.
Principal scope error
The most common failure mode: the principal authorised a scope that was too broad to have meaningfully understood, and the agent acted within it producing an unwanted outcome.
Marketplace protocol error
Failure mode where the marketplace's rules allowed a transaction that should not have been allowable — e.g. failing to require a confirmation step that would have caught the error.
EU AI Act
The European Union's horizontal AI regulation. The original 2024 framework was already obsolete on agentic systems by enactment; the agentic-system amendments are arriving faster than industry expects.
Moral crumple zone
The pattern of assigning blame to the human nearest the failure of an automated system. Agentic AI compounds the problem because the chain of authorisation is longer.
Customer experience economics
The dual-sided dynamics of customer service and the structural reasons agentic CX is the only path through them.
Asymmetric service economy
The dynamic where customer-service spending grows year-over-year while customer satisfaction declines. The structural reason customer support is broken on both sides.
Deflection
Customer-service tooling pattern that reduces ticket volume without resolving underlying issues. Cost-shifting, not resolution.
Per-contact cost
The unit economic of a customer-support interaction. Has risen every year of the past decade with no offsetting gain in resolution quality.
Personalisation at scale
The capacity to deliver per-customer reasoning across high volumes — the conversations where the company has the most to gain are the ones least served by either economic option.
Engineered hostility
Customer-service patterns deliberately designed to discourage contact (complex IVRs, hidden contact info, infinite chatbot loops). Telecom is the textbook case.
Agentic CX
Customer experience powered by agentic AI rather than chatbots or human agents alone. The category that's emerging where Intercom Fin, Sierra, Decagon, and Teleperson all play.
Pricing & business models
How agentic platforms get paid — and why per-seat SaaS doesn't survive the transition.
Per-seat SaaS
The pricing model built for software humans use. Agentic AI doesn't have seats; the next decade's pricing model looks different.
Outcome-based pricing
Charging per resolved ticket, recovered charge, or completed task — rather than per user-seat or per token. Aligns vendor incentives with customer value.
Hybrid pricing
Combination of subscription + outcome-based + brand-funded marketplace fees. The likely steady state for consumer-side agentic advocates.
Brand-funded marketplace pricing
Pricing model where the consumer pays nothing and the marketplace is funded by the brands paying for discoverability. Highest engagement, biggest alignment risk.
Teleperson product vocabulary
Product surfaces and concepts unique to Atlas, Teleperson's consumer assistant, and the brand-side platform.
Atlas
Teleperson's AI assistant for Mac. It calls companies for you, always tells them it's an AI, waits on hold, remembers every call, and checks your bank to make sure companies do what they promised.
Vendor Hub
Atlas's directory of every company you pay, built from your bank activity: spend, subscriptions, phone menus and the closest store.
Follow-through
Atlas tracking a promise made on a call — a refund, cancellation or lower price — against your bank until it's kept, and telling you when it isn't.
Dispute pack
What Atlas assembles when you're double charged or a price jumps: the charges, what the company said and when, and ready-to-send drafts for the company and your bank.
AI Agent (brand-side)
The branded chat-and-voice agent a company adds to its own site with one script tag. It recognizes the customer, answers from the company's knowledge base, and hands off to a human when judgment or law requires it.
Term missing, or wrong?
The glossary is curated from the Research Library and the language we hear in design-partner conversations. Send us a note and we'll iterate.