1. Home
  2. Resources
  3. Glossary

AI, ERP and CRM glossary, from Agentforce to webhooks

MCP, short for Model Context Protocol, is a shared standard that lets AI apps such as Claude and ChatGPT connect to other software. This page explains it and 27 more AI, ERP and CRM terms in plain words. Each term starts with a short answer. Then come the detail, the primary source we read on 15 or 16 November 2026, and where the term shows up in the systems we work in.

28 termsChecked 16 November 2026Every source linked

Agentforce

Agentforce is Salesforce's own tool for building AI agents that work inside Salesforce. Salesforce calls it an AI agent platform. Teams use it to build, test, launch and watch over their agents in one place.

Salesforce says Agentforce is part of the Salesforce Platform. Your existing workflows, prompt templates, code and APIs can become steps an agent takes. For data safety, Salesforce points to its Einstein Trust Layer. It lists zero data retention among its features. An audit trail shows what each agent did.

Our team has built AI agents in Salesforce. Agentforce is one route in. A custom agent that reaches Salesforce through its API is another.

Source: Salesforce, How Does Agentforce Work?, read 16 Nov 2026

Agentic AI

Agentic AI is AI that works through a job step by step, using tools, until the job is done. Anthropic calls them “agentic systems” and splits them into two kinds. In workflows, code sets each step. In agents, the AI model picks its own steps and tools.

Anthropic says workflows are more predictable for clear, well-defined tasks. Agents suit work where the model has to make the decisions. Its advice is to start with the simplest build, and add agent steps only when that falls short.

Source: Anthropic, Building effective agents, published 19 December 2024, read 15 Nov 2026

AI agent

An AI agent is a program that uses an AI model to finish a job by itself, one step at a time. It picks a step and uses a tool, such as a system's API. Then it checks the result and repeats until the job is done or it needs a person.

Anthropic's engineers describe agents as “LLMs using tools based on environmental feedback in a loop”. The same article warns that agents cost more, and that one mistake can feed the next. It advises plenty of testing in a sandbox, a safe copy of the system, first.

An agent in an ERP can save a draft and wait, so a person approves each record, such as a quote or an order, before it is sent or posted.

Source: Anthropic, Building effective agents, published 19 December 2024, read 15 Nov 2026

AI governance

AI governance is the set of rules a company uses to decide how it uses AI and who answers for the results. It covers what may be built or bought, how risks are checked, and who signs off.

In the US, one voluntary guide to it is the AI Risk Management Framework 1.0 (NIST AI 100-1). NIST, a US government agency, published it in January 2023. It names four functions: Govern, Map, Measure and Manage. NIST describes Govern as running through the other three.

Source: NIST, AI Risk Management Framework 1.0 (NIST AI 100-1), January 2023, PDF, read 15 Nov 2026

On our siteAI governance consulting

API (application programming interface)

An API is a set of rules that lets one program use another program directly, with no person clicking through its screens. MDN, Mozilla's guide for web developers, calls it a contract between the program that offers it and the software that uses it.

NetSuite's API for new work is its REST web services. SAP Business One's is the Service Layer, its web interface for other software. An AI agent reaches a system through an API like these, or through an MCP server built on one.

Sources: MDN Web Docs, Glossary: API, read 15 Nov 2026; Oracle NetSuite help, SOAP Removal Plans FAQ, read 15 Nov 2026; SAP Business One, Service Layer API Reference, read 15 Nov 2026

Breeze Assistant (HubSpot)

Breeze Assistant is HubSpot's built-in AI chat assistant. You ask it something, and it answers from what your HubSpot account holds, such as the CRM record you ask about. HubSpot calls it a conversational AI assistant.

What it can see and do depends on the user's own HubSpot permissions. HubSpot says Breeze Assistant does not use HubSpot Credits, though an AI agent run it starts may cost credits.

When we read HubSpot's AI product page on 16 November 2026, it was titled Agent Hub. It lists ready-made agents, such as a customer agent and a data agent, and an agent builder for custom agents. HubSpot says custom agents built there use HubSpot Credits. Our team has built AI agents in HubSpot.

Sources: HubSpot Knowledge Base, Understand Breeze Assistant, read 16 Nov 2026; HubSpot, Agent Hub, read 16 Nov 2026

Business process mapping

Business process mapping means drawing how a piece of work gets done today, step by step. The map shows who does each step, what goes in and out, and where someone makes a decision. ASQ, the American Society for Quality, calls a flowchart a common type of process map.

ASQ's advice is that the people who do the work should draw the map, not a technical expert. The first map shows how things run now. Then the team looks for delays, rework, duplicate steps and too many handoffs.

BPMN, short for Business Process Model and Notation, is a standard set of symbols for these maps. The Object Management Group published version 2.0.2 in January 2014. It calls BPMN the de facto standard for business process diagrams. In ERP consulting, the map of today's work shows what the new system has to cover.

Sources: ASQ, What is a flowchart?, read 16 Nov 2026; Object Management Group, About BPMN 2.0.2, read 16 Nov 2026

CPQ (configure, price, quote)

CPQ is software that helps a sales team put together a correct quote. The letters stand for configure, price, quote. Salesforce says it helps teams set up products, apply pricing rules, handle discounts and approvals, and produce quotes.

Salesforce places CPQ at the start of quote-to-cash, the full path from a quote to a contract, a bill and a payment. CPQ covers the configure, price and quote steps. A CPQ tool prices what a salesperson enters. RFQ automation starts one step earlier, with the request a customer emails in.

Source: Salesforce, What is CPQ, or Configure, Price, Quote?, read 16 Nov 2026

CRM (customer relationship management)

A CRM is the system a company uses to keep track of its customers and sales deals. Contacts, deals, emails and calls sit in one shared record. Salesforce, which sells one, describes CRM as technology for managing a company's relationships and interactions with customers.

Our team has built AI agents in two CRMs, Salesforce and HubSpot. A CRM holds the customer side, and an ERP holds orders and invoices. So an agent that turns a won deal into a sales order needs a way into both.

Source: Salesforce, What is CRM?, read 15 Nov 2026

On our siteAI agents for SalesforceAI agents for HubSpotAI for CRMCRM implementation

ERP (enterprise resource planning)

An ERP is the main software a company runs its orders, stock and accounts in, kept in one shared set of records. SAP describes ERP as software for a company's core business processes that gives “a single source of truth”.

SAP's own list of those processes runs from finance and HR to manufacturing, supply chain, sales and buying.

Our team has taken ERPs live on SAP, Oracle NetSuite, Odoo, Microsoft Dynamics, ERPNext, Xero, QuickBooks, Sage and custom ERPs. It has also built agents that draft quotes, bookkeeping entries and orders in an ERP.

Source: SAP, What is ERP?, read 15 Nov 2026

Generative AI

Generative AI is AI that makes new things, such as text, pictures, sound or code. It learns patterns from the data it was trained on, and large language models such as Claude are one kind.

NIST's Generative AI Profile (NIST AI 600-1, July 2024) lists 12 risks that generative AI creates or makes worse. Number 2 on the list is confabulation, which people often call hallucination.

Source: NIST, Generative Artificial Intelligence Profile (NIST AI 600-1), July 2024, PDF, read 15 Nov 2026

On our siteGenerative AI development

Go-live

Go-live is the day a new system starts doing real work, such as an ERP that starts taking real orders. Microsoft calls it a critical milestone: after a successful go-live, the new solution is operational.

Microsoft's guide for Dynamics 365 lists what should be done first. Testing is signed off. That includes user acceptance testing, where staff try the system on their own tasks. There is a plan to move the data, and a cutover plan for the switch from the old system. Users are trained and given the right security roles. A support plan is signed off.

Microsoft also warns that after go-live, links to other systems and bulk jobs can cause peaks in use that hit the system's API limits.

Source: Microsoft Learn, Dynamics 365 implementation guide: Prepare to go live, read 16 Nov 2026

Hallucination

A hallucination is when an AI model says something false as if it were true. NIST's formal name for it is confabulation.

NIST says it comes from the way these models are built. They predict the likely next word, and that is not the same as checking a fact. So an agent that works in an ERP should show the documents it used, and wait for a person before anything is posted.

Source: NIST, Generative Artificial Intelligence Profile (NIST AI 600-1), July 2024, PDF, read 15 Nov 2026

On our siteRAG development

Human in the loop (HITL)

Human in the loop means a person checks the AI's work and approves or rejects it before it takes effect. In an AI agent, the run pauses at that step until someone decides.

Some ERPs build this in. In SAP Business One, approval procedures can be switched on. Then an order that outside software posts through the Service Layer becomes a draft, and SAP returns error code -2028. The draft waits for an approver inside Business One.

Sources: OpenAI Agents SDK, Human-in-the-loop, read 15 Nov 2026; SAP Business One, Service Layer API Reference, read 15 Nov 2026

Intuit Assist and Intuit Intelligence

Intuit Assist is the AI helper Intuit made for its products, QuickBooks among them. It is built on generative AI. QuickBooks help files its article on Intuit Intelligence under Intuit Assist. Intuit says that tool answers questions and does money tasks in QuickBooks, using your own company data.

Intuit's article was updated on 21 October 2026. It says the tool works in QuickBooks Online and Intuit Accountant Suite. The chat comes with your QuickBooks plan, but prompts have a monthly limit. A $10 a month add-on raises the limit. No plan makes it unlimited.

Intuit's launch page says Intuit Assist would help in TurboTax, Credit Karma and QuickBooks. Our team's ERP go-lives include QuickBooks.

Sources: QuickBooks help, Introducing Intuit Intelligence, updated 21 October 2026, read 16 Nov 2026; Intuit, Intuit Assist, read 16 Nov 2026

Joule (SAP)

Joule is SAP's AI, built into SAP's cloud software. SAP names four parts. Joule Work answers business questions. Joule Assistants and Joule Agents run workflows. Joule Studio is where you build AI tools of your own.

SAP describes Joule Agents as ready-to-use AI agents. They pick from tools, such as Joule skills, other agents and outside apps, to carry out a plan. SAP says customers with SAP cloud subscriptions can switch on Joule base for free.

SAP's integration guide, dated 4 November 2026, says Joule runs on SAP BTP, a cloud platform from SAP. We searched its 236 pages for “Business One” and found nothing. Our research note on SAP Business One and Joule has the details.

Sources: SAP, Joule Business AI Solutions, read 16 Nov 2026; SAP, Joule Assistants, read 16 Nov 2026; SAP Help, Integrating Joule with SAP Solutions, 4 November 2026, PDF, read 16 Nov 2026

LLM (large language model)

A large language model, or LLM, is an AI model trained on a huge amount of text. It can answer questions, sum up documents and write. Anthropic's Claude is built on one.

NIST explains that an LLM writes by predicting the next word, or part of a word. On its own it has no live link to your systems. Tool calling and MCP give it one.

Sources: Claude Platform Docs, Glossary, read 15 Nov 2026; NIST, Generative Artificial Intelligence Profile (NIST AI 600-1), July 2024, PDF, read 15 Nov 2026

On our siteLLM development services

MCP (Model Context Protocol)

MCP, short for Model Context Protocol, is a standard way for AI apps to connect to other software. The MCP project calls it an open-source standard. An AI app such as Claude or ChatGPT uses it to read data in a system and act in it. It does this through a program called an MCP server.

The project compares MCP to a USB-C port for AI apps: one standard plug instead of a new cable for every pair. Its site lists the version dated 28 July 2026 as the latest.

Three ERPs we work in now offer an MCP route. Oracle's NetSuite AI Connector Service works with MCP clients such as Claude and ChatGPT. Odoo 20 can act as an MCP server that reads or changes records. SAP added an open-source MCP server for the Business One Service Layer in feature package 2608.

Sources: Model Context Protocol, What is MCP?, read 15 Nov 2026; Model Context Protocol, Architecture overview, read 15 Nov 2026; Oracle NetSuite help, NetSuite AI Connector Service FAQ, read 15 Nov 2026; Odoo 20 documentation, AI MCP server, read 15 Nov 2026; SAP Business One 10.0 FP 2608, What's New: Platform and Extensibility, read 15 Nov 2026

On our siteMCP server developmentNetSuite AI agents

Multi-agent system

A multi-agent system splits one job between several AI agents, each with its own part. Anthropic defines it as multiple agents, each an AI model using tools in a loop, working together.

In Anthropic's research system, a lead agent plans the work and hands parts to subagents that run at the same time. The cost is higher. In Anthropic's data, agents used about 4 times as many tokens as a chat, and multi-agent systems about 15 times. Tokens are the pieces of words a model reads and writes.

Our team built a multi-agent system that automates RFQs and RFPs end to end, on both sides. A customer's request becomes a quote, and RFQs go out to suppliers.

Source: Anthropic, How we built our multi-agent research system, published 13 June 2025, read 16 Nov 2026

OAuth 2.0

OAuth 2.0 is a standard way to let an app into a system without handing over your password. The app gets an access token, a kind of digital key that opens only what was approved.

The IETF (Internet Engineering Task Force) published it as RFC 6749 in October 2012. NetSuite now points new integrations to it. Oracle says that from release 2027.1, no new integration can use token-based authentication, the older sign-in method.

Sources: IETF, RFC 6749: The OAuth 2.0 Authorization Framework, October 2012, read 15 Nov 2026; Oracle NetSuite help, Preparing for Token-based Authentication (TBA) End of Support, read 15 Nov 2026

RAG (retrieval-augmented generation)

RAG lets an AI model look things up before it answers. Short for retrieval-augmented generation, it works in two steps. Software first finds the documents that match a question. Then it gives them to the model along with the question. The model can then answer from facts it never saw in training, such as your own price list.

The name comes from a 2020 research paper by Patrick Lewis and 11 co-authors, accepted at the NeurIPS 2020 conference. In that paper the documents came from Wikipedia. Anthropic notes that RAG is only as good as the documents it finds.

Sources: Lewis and others, Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, arXiv 2005.11401, read 15 Nov 2026; Claude Platform Docs, Glossary, read 15 Nov 2026

On our siteRAG development services

REST API

A REST API is an API that works over the web. Software reads or changes a record by sending a request to a web address. MDN says beginners can take a REST API to mean an HTTP service that standard web tools can call. HTTP is the set of rules the web runs on.

REST, short for Representational State Transfer, comes from chapter 5 of Roy Fielding's dissertation of 2000. Oracle says that from release 2026.1, all newly built NetSuite integrations should use REST web services with OAuth 2.0.

Sources: MDN Web Docs, Glossary: REST, read 15 Nov 2026; Roy Fielding, dissertation, chapter 5: Representational State Transfer, 2000, read 15 Nov 2026; Oracle NetSuite help, SOAP Removal Plans FAQ, read 15 Nov 2026

RFP (request for proposal)

An RFP, or request for proposal, is a document a buyer sends to ask suppliers how they would do a job and on what terms. Where an RFQ asks for prices, an RFP asks for a full proposal. US federal buying rules list what an RFP for a competitive purchase must cover. It names the need and the expected contract terms. It also says what each proposal must contain and how proposals will be judged.

Those rules are the Federal Acquisition Regulation, or FAR, section 15.203. It says RFPs are used in negotiated purchases, and that they may be sent and answered electronically.

Source: Acquisition.gov, FAR 15.203 Requests for proposals (FAC 2026-01, effective 13 March 2026), read 16 Nov 2026

RFQ (request for quotation)

An RFQ, or request for quotation, is a document a company sends to suppliers to ask for their prices. In Odoo, once the vendor agrees, the RFQ is confirmed as a purchase order.

NetSuite keeps RFQs as records too. Each one sets the item, its price tiers and a bidding window with an open and a close date. Once every vendor has quoted, your buyer compares the quotes to find the best overall deal.

An RFQ can arrive by email. An AI agent can read it and prepare the matching record in the ERP for a person to check.

Sources: Odoo 20 documentation, Requests for quotation, read 15 Nov 2026; Oracle NetSuite help, Request for Quote, read 16 Nov 2026

On our siteRFQ and RFP automationRFQ and quote automation

Sage Copilot

Sage Copilot is the AI helper built into some Sage products, such as Sage Intacct. Sage calls it a generative AI assistant. It helps with finance tasks and looks into your business data.

Sage says it helps close the books at month end, flags odd items in the books and helps process invoices. In Sage X3, it helps with day-to-day tasks. On Sage's AI page, called Sage Ai, it is the AI business assistant.

It is not in every Sage product or every country. Sage also warns that some features it shows are plans. It says not to rely on future features when you buy.

Sources: Sage US, Sage Copilot, read 16 Nov 2026; Sage US, Sage Ai, read 16 Nov 2026

Sandbox

A sandbox is a practice copy of a real system, so you can build and test without touching real records. Oracle says a NetSuite sandbox has the same setup, data and customizations as the live account.

A NetSuite sandbox keeps the live account's ID with _SB1, _SB2 and so on added to the end. Anthropic's engineers advise plenty of testing of AI agents in sandboxed environments before they run for real.

Sources: Oracle NetSuite help, NetSuite Sandbox, read 15 Nov 2026; Oracle NetSuite help, About Sandbox Accounts, read 15 Nov 2026; Anthropic, Building effective agents, published 19 December 2024, read 15 Nov 2026

Tool calling

Tool calling lets an AI model ask your software to do a task for it, such as looking up an order. It is also called function calling. The model sends a structured request, your code runs it, and the result goes back to the model.

The model only asks, and your code decides what runs. Vendors set limits too. Oracle says the custom tools that AI clients call through NetSuite's AI Connector Service cannot use the N/http, N/https, N/llm or N/sftp modules.

Sources: Claude Platform Docs, Tool use with Claude, read 15 Nov 2026; Oracle NetSuite help, SuiteScript 2.1 Custom Tool Script Type, read 15 Nov 2026

Webhook

A webhook is a message one system sends to another the moment something happens, such as a new order. The other system does not have to keep asking whether anything changed.

SAP Business One's Service Layer sends webhooks. Feature package 2608 added conditions, so a webhook can fire only when, for example, a business partner's address changes. By default a message carries only the key, or ID, of the changed record.

Sources: GitHub Docs, About webhooks, read 15 Nov 2026; SAP Business One 10.0 FP 2608, What's New: Platform and Extensibility, read 15 Nov 2026

Tell us the job and the system it runs in

Name the system you run and the work you have in mind. A person on our team replies by email. The first call is free.

Tell us the job

Questions about these terms

What is the full meaning of MCP?

In AI, MCP stands for Model Context Protocol. The MCP project calls it an open-source standard for connecting AI applications to external systems, and compares it to a USB-C port for AI apps. The same three letters mean other things in other fields. This glossary covers only the AI meaning.

How is MCP different from an API?

An API is the way into one system, with that system's own rules. MCP is one shared standard for how AI apps find and use tools across many systems. An MCP server often calls the system's API underneath. Oracle says NetSuite's MCP Standard Tools SuiteApp needs the REST Web Services permission for some of its tools.

Does ChatGPT use MCP?

Yes. The MCP project lists ChatGPT and Claude among the AI apps that support it, along with developer tools such as Visual Studio Code and Cursor. Oracle's NetSuite FAQ adds that some ChatGPT plans need Developer Mode to use MCP connectors. We read both on 15 November 2026.

What is an LLM, and what is MCP?

An LLM, or large language model, is the AI model that reads and writes text. Anthropic's Claude is built on one. MCP is the standard an AI app uses to connect that model to other software. Anthropic's glossary describes MCP as an open protocol that standardizes how applications provide context to LLMs.

What is the difference between RAG and an LLM?

The LLM is the model that writes the answer. RAG is a way of feeding it: software finds the right documents first and passes them in with the question. An LLM can run without RAG, but RAG always needs a language model to write the answer.

What is the difference between agentic AI and an AI agent?

An AI agent is one kind of agentic AI. Anthropic uses the name “agentic systems” for two kinds. In agents, the model picks its own steps and tools. In workflows, code sets each step. So every AI agent is agentic AI, but a fixed workflow can count as agentic AI with no agent in it. Our article on agentic AI vs AI agents goes further.

What is the difference between generative AI and agentic AI?

Generative AI makes content, such as text, pictures, sound or code. Agentic AI uses an AI model to carry out a task step by step, with tools. The two often work together. Anthropic describes agents as LLMs using tools in a loop, and an LLM is one kind of generative AI.

What are the components of an AI agent?

A language model, the tools it can call, and a loop that joins them. Anthropic's engineers describe agents as LLMs using tools in a loop. In the loop the model acts, checks the result and picks the next step. Anthropic calls the basic building block an LLM with extras such as retrieval, tools and memory. A person can also be added as a checkpoint. That is human in the loop.

What is HITL in AI?

HITL stands for human in the loop. It means a person reviews the AI's work and approves or rejects it before it takes effect. OpenAI's Agents SDK, a toolkit for building agents, uses it to pause a run until a person approves or rejects a sensitive tool call. In SAP Business One, approval procedures do the same job for orders that outside software posts.

What are the risks of using AI agents?

Anthropic's engineers name two: higher costs, and errors that compound when one wrong step feeds the next. NIST adds confabulation, where a model states false things with confidence. Anthropic advises testing in sandboxed environments with guardrails. A limited login for each agent, and a person who approves each record before it is posted, cut the risk further.

What is the difference between an RFQ and an RFP?

An RFQ asks suppliers for prices. An RFP asks for a full proposal: how they would do the job and on what terms. In Odoo, a confirmed RFQ becomes a purchase order. US federal rules say an RFP for a competitive purchase must also say how proposals will be judged. Our team built a multi-agent system that handles both, on both sides.

What is Agentforce?

Agentforce is Salesforce's tool for building AI agents inside Salesforce. Salesforce says teams use it to build, test, launch and watch over their agents. It is part of the Salesforce Platform, so your existing workflows, code and APIs can become steps an agent takes. Our team has built AI agents in Salesforce. The Agentforce entry above has more.

What are the pillars of AI governance?

There is no single official list. NIST, a US government agency, published a voluntary AI Risk Management Framework in January 2023. It uses four functions: Govern, Map, Measure and Manage. NIST describes Govern as running through the other three.