Can I see what the Agent will own before I buy?
Loading AI Team page
Use this glossary to decode setup, access, approval, cost, proof, and Agent OS language before you choose a workflow or approve go-live.
Buying clarity
You do not need to become technical. You do need to know what the Agent owns, what it can access, what it may do alone, and what proof you should receive.
Can I see what the Agent will own before I buy?
Do I know which actions need approval?
Do I understand which provider costs can pass through?
Can I revoke access if the workflow changes?
Will I see completed work, exceptions, and useful reports after go-live?
Do I know when AI Team pauses, escalates, or asks to re-scope?
Definitions
Use these terms when deciding whether an Agent is the right way to remove a recurring task.
6 buyer terms
An AI Agent handles a defined workflow by reading context, preparing work, updating systems, and asking for approval when rules require it.
Why it matters: You buy a repeatable business result, not a generic chat tool or another employee to manage.
Agent catalogA managed AI Agent is configured, monitored, improved, and supervised by AI Team instead of being left for your team to operate.
Why it matters: Your team gets the work output without becoming responsible for prompts, model routing, QA, and exception handling.
Human supervisionService-as-Software means software performs a service outcome under managed rules, instead of giving you another software tool to configure alone.
Why it matters: The buying question becomes whether the work gets removed from your team, not whether the tool has many features.
About AI TeamThe Agent catalog is the menu of defined workflows AI Team can scope, price, set up, and operate for buyers.
Why it matters: You can start with one specific task instead of commissioning a broad automation project.
Browse AgentsA Standard Agent handles lower-risk, repeatable work with clear rules, modest volume, and limited connected-system complexity.
Why it matters: This is usually the lowest monthly price band when the task is narrow enough to operate predictably.
Pricing bandsA Managed / Custom Agent covers work where volume, regulation, finance impact, customer risk, or integration complexity changes the operating burden.
Why it matters: Custom pricing prevents a risky workflow from being under-scoped and then failing after go-live.
Pricing bandsDefinitions
These terms decide whether an Agent can safely move from interest to live work.
6 buyer terms
Setup review confirms the workflow, source systems, required access, success criteria, approval needs, and known blockers before build work starts.
Why it matters: It prevents a vague request from becoming an Agent that cannot be judged or safely launched.
Setup processAccess review identifies which accounts, roles, OAuth grants, API keys, files, and permissions the Agent needs before live work starts.
Why it matters: You know what the Agent can touch, who owns access, and how access can be removed.
Credential governanceScope confirmation writes down the work included, the work excluded, the success metrics, and the situations that require a pause or re-scope.
Why it matters: It protects both sides from an Agent being treated like an unlimited employee.
Service policyAn approval rule says which actions the Agent may complete automatically and which decisions need a human before anything is sent, changed, or spent.
Why it matters: Customer-facing, financial, public, sensitive, and high-cost actions stay controlled.
Deployment readinessDeployment QA tests realistic inputs, missing data, permission failures, vendor outages, cost limits, and escalation paths before go-live.
Why it matters: The Agent should prove it can handle imperfect conditions before it handles live work.
Deployment readinessThe go-live gate is the final approval point where scope, access, costs, approval rules, dashboard metrics, QA evidence, and operator coverage are checked.
Why it matters: An Agent should not start live work simply because checkout or setup intake happened.
Deployment readinessDefinitions
These terms help buyers avoid surprise bills and unsafe access handoffs.
6 buyer terms
A client-owned API key is a provider key that belongs to the buyer's account and is used only after the required scope and cost rules are approved.
Why it matters: You keep ownership of paid provider accounts where practical and can revoke access when the work changes.
API keys and costsAn OAuth grant lets an approved app access a service without sharing a raw password.
Why it matters: It gives the buyer a clearer permission and revocation path than sending credentials through email or forms.
Credential governanceA connected system is a CRM, inbox, store, support desk, accounting tool, spreadsheet, ad account, or other work system the Agent needs to read or update.
Why it matters: The number and risk level of connected systems affect setup complexity, price band, and go-live readiness.
IntegrationsA pass-through cost is a third-party provider charge, such as voice, enrichment, scraping, messaging, ads, SEO data, or premium API usage, that is billed or approved separately.
Why it matters: You can approve expensive provider usage before the Agent creates a bill.
API keys and costsA spend cap is the approved limit where paid provider usage must stop, warn, or wait for a human decision.
Why it matters: The Agent should pause before usage turns into an unexpected invoice.
Billing governanceCredential revocation removes or disables the Agent's access to a tool, account, API key, OAuth grant, or service account.
Why it matters: You need a clean off-switch when a provider changes, an employee leaves, a project ends, or a security concern appears.
Credential governanceDefinitions
These terms explain what buyers should see once an Agent is running.
6 buyer terms
A work packet is the buyer-visible output from an Agent, such as a draft reply, triage summary, checklist, research brief, cleanup report, or exception note.
Why it matters: You can judge whether the Agent is producing useful work instead of only seeing activity counts.
Work samplesA completed work feed shows the work the Agent finished, the work waiting for approval, and the work blocked by missing access or unclear rules.
Why it matters: You get visibility without reading raw prompts, private logs, or provider internals.
Operational proofA run log records why an Agent started work, what it did, what it changed, where it paused, and which approval or exception state applies.
Why it matters: Run logs support review, troubleshooting, cost checks, and handoff when something needs human attention.
Operational proofAn exception is a blocked, risky, unclear, failed, or out-of-scope situation the Agent should not force through automatically.
Why it matters: Exceptions stop unreliable output from becoming customer-facing or business-critical action.
SupportEscalation routes an exception, approval, incident, complaint, access issue, or sensitive decision to the right human owner.
Why it matters: Your team and AI Team know who must decide before the workflow continues.
Human supervisionHuman supervision means AI Team reviews quality, exceptions, approvals, incidents, setup changes, and workflow improvements behind the scenes.
Why it matters: Automation keeps moving routine work while humans handle judgment, risk, and improvement.
Human supervisionDefinitions
These terms explain the operating layer without forcing buyers to become technical operators.
6 buyer terms
Agent OS is the AI Team operating layer for setup state, approvals, connected tools, runs, exceptions, costs, QA, release control, and reporting.
Why it matters: It keeps many client Agents controlled in one operating system instead of many unmanaged automations.
Agent OSModel routing selects the AI model route for a task based on quality, cost, privacy, latency, fallback rules, and approved provider boundaries.
Why it matters: AI Team can lower AI cost where quality is proven and use stronger routes when the work needs them.
Model routingA workhorse model is the default lower-cost model route used for standard work after it passes AI Team quality, tool-use, and safety checks.
Why it matters: Lower model cost helps keep monthly Agent pricing attractive without removing quality gates.
Model routingA fallback model is a backup route used when the default route is unavailable, not strong enough, too slow, or unsuitable for the workflow.
Why it matters: The Agent can preserve quality and reliability instead of failing silently or using the wrong model.
Model routingA tool call is a controlled action where an Agent reads or updates an approved system, requests data, or triggers a workflow step.
Why it matters: Tool calls need boundaries because they can affect records, customers, costs, or downstream teams.
Deployment readinessSource grounding means the Agent uses approved source material, connected systems, or cited references instead of guessing from memory.
Why it matters: Grounded output is easier to check, approve, and correct when source data is missing or wrong.
Work samplesRelated pages
These pages explain how the terms become concrete before an Agent goes live.
FAQ
Direct answers for buyers who want plain-English meaning without reading technical documentation.
Buyers should understand setup, access, approval, cost, proof, and supervision terms before they hire an Agent. Plain definitions reduce confusion during scope review.
No. These are buyer-facing explanations. Signed order forms, terms, the DPA, retention policy, and service policy control final obligations.
No. Buyers should understand the terms that affect scope, cost, access, approval, risk, reporting, and support. AI Team handles the technical operation behind the scenes.
Scope confirmation, access review, approval rule, pass-through cost, spend cap, deployment QA, go-live gate, exception, escalation, and human supervision matter most before live work starts.