1. Home
  2. Services
  3. AI agent development

AI agent development for the jobs your team does every day

We are an AI agent development company for teams that run on an ERP, the software that holds your orders and accounts, or a CRM, where you keep customers and deals. An AI agent is a program that does one job. It reads a document, such as an emailed request for a quote, and checks it against your rules. Then it saves a draft in NetSuite, Odoo, SAP, Salesforce, HubSpot or your own ERP, and a person on your team usually approves it.

RFQs and RFPs, both sidesDaily bookkeepingPurchase and sales ordersSalesforce and HubSpot agents
Agent run, quote requests
Quote QT-4471
Draft
quote_request_1182.eml Read
4 of 4 lines matched to the item list Matched
Prices from the customer's price list Applied
Saved as a draft for the sales desk

Illustration with sample data.

What an AI agent development company builds

One AI agent that does one job in one system. Or a few agents that share a bigger job, such as handling quotes from the first email to the last reply. The work starts when something arrives, such as a request for a quote by email. It ends with a draft record in your ERP or CRM, such as a quote or a sales order, for a person to approve.

We build custom AI agents for business. A build can include tests on your past documents, a login of its own for the agent and support after launch. What each build includes is agreed per project. Our team has taken ERPs live on SAP, Oracle NetSuite, Odoo, Microsoft Dynamics, ERPNext, Xero, QuickBooks, Sage and custom ERPs. We have also built AI agents in Salesforce and HubSpot.

Which page is for you

We have three pages on building AI. Pick the one that sounds like your job.

AI agent development, this page

You have one job that repeats, such as typing emailed orders into your ERP. You want a program that does it and hands each draft to a person.

Agentic AI development

Your job has many steps that change from case to case. You need to decide which steps the AI may choose and where it must stop.

AI development services

You want AI software built and have not picked its shape yet. That page covers the whole build and what it costs to run.

Not sure yet

Start with AI consulting services to pick the first job, or see every service we offer.

Four kinds of AI agents our team has built

Most start with a document that arrives by email and end with a draft in the ERP or CRM. Clients are not named on this site, and we give no numbers from their projects.

RFQs and RFPs, on both sides

A system of several agents. On the sales side, a customer's request becomes a draft quote. On the buying side, requests for quotes go out to suppliers and their replies come back in.

Request by email
Draft quote or supplier RFQ For a buyer or the sales desk

Daily bookkeeping

The agent drafts the day's bookkeeping entries from the documents that came in. Someone in finance checks them before they post.

The day's documents
Draft ledger entries For finance

Purchase and sales orders

The agent turns an order document into a draft purchase order or sales order, with the right items and prices.

Order document
Draft order in the ERP For approval

Agents in Salesforce and HubSpot

Agents that work on CRM records, the customers and deals your sales team keeps. In HubSpot, our team has built agents on Claude, the AI model made by Anthropic.

CRM record or email
Updated record or draft For the record's owner

More on three of these: RFQ and RFP automation, AI agents for Salesforce and AI agents for HubSpot.

A worked example: RFQs and RFPs on both sides

Our team built a system of several AI agents that runs RFQs and RFPs end to end. An RFQ is a request for a quote. An RFP is a request for a proposal, which asks for a plan as well as a price. The table shows how that kind of work splits between agents. Which agents a build has, and what each may do, is agreed per project.

Scroll the table sideways to see every column.

SideWhat arrivesWhat an agent doesWhat a person checks
SalesA customer's RFQ by email, often with a PDF or a spreadsheet of linesReads each line and matches it to your item list. A line with no match can be flagged for a person.The lines it could not match
SalesThe matched linesApplies the customer's price list and terms, then saves a draft quote in the ERP or CRMThe sales desk approves the quote before it is sent
SalesA customer's RFP, with questions to answerDrafts answers from what you have approved before, and lists the questions it cannot answerThe person who signs the proposal
Buying
Second side
A need to buy, such as items short for an orderDrafts an RFQ to the suppliers you name for those itemsA buyer approves the RFQ before it goes out
BuyingEach supplier's reply, often a PDFReads the price, the quantity, the delivery date and how long the offer is valid, and saves them against the RFQA buyer compares the offers side by side and picks one
Where the agents write

Many ERPs already have records for both sides. NetSuite has a Request for Quote feature for vendors. Odoo 20 Purchase creates and sends RFQs. ERPNext has a Request for Quotation, and a Supplier Quotation that keeps each offer in set fields, so it can become a purchase order without retyping. Read 16 November 2026.

Check the vendor's own agent first

Microsoft's Sales Order Agent in Business Central reads customer requests sent by email and prepares sales quotes. Microsoft says it always involves named users to review and approve all outgoing messages before they go to customers. Read 16 November 2026.

The full page on this work is RFQ and RFP automation.

How a custom AI agent connects to your system

Four parts, in this order. The way into each ERP or CRM is different, so we plan it from the vendor's own documents.

Input

The document arrives

An email, a PDF or a file in a folder your team already uses.

Agent

The agent reads it

It pulls out the lines, checks them against your item list and price lists, and applies your rules.

API

It writes through the system's API

An API is the door a system opens for other software. The agent uses the one your ERP supports.

Person

Your team checks it

The draft waits in the ERP for the person who owns that step.

NetSuite

Oracle says that from release 2026.1, new integrations should use REST web services with OAuth 2.0, a standard way to sign in. Read 15 November 2026.

SAP Business One

The Service Layer, SAP's API, can create orders, quotations and drafts. Feature package 2608 added an MCP server for AI agents. MCP, short for Model Context Protocol, is a standard way for an AI tool to use a system's functions. Read 15 November 2026.

Odoo

Odoo 20 has an MCP server, so an outside agent can read or change records. The outside API needs Odoo's Custom plan, and its keys last three months at most. Read 15 November 2026.

Custom ERPs

No vendor API is needed. The agent can work through the database on a role of its own. PostgreSQL and SQL Server ship read-only roles, and PostgreSQL's row security limits which rows a user can add or change. Read 9 October and 15 November 2026.

Salesforce

An integration user with the Minimum Access, API Only profile. Salesforce suggests one such user for every integration, so each one sees only its own share of data. Read 16 November 2026.

HubSpot

The remote HubSpot MCP server lets an outside agent read, write and update records over OAuth. HubSpot's own connectors for Claude, ChatGPT, Copilot and Gemini run on it. Read 16 November 2026.

To build an agent yourself, our guide on how to build an AI agent for an ERP lists the seven steps.

Systems we build AI agents in

Pages with a link are live. Each one is written from the vendor's own documents.

NetSuiteSAPSAP Business OneOdooERPNextSageQuickBooksCustom ERPsSalesforceHubSpot

How our AI agent development services usually run

Four steps. What each step covers, and how long it takes, is agreed per project.

Step 1

Read the system

We read your ERP's version, the vendor's documents for it, and the emails or files that start the job.

Step 2

Agree the rules

Which price list, which warehouse, which customer terms, which approver. The agent applies the rules you sign off.

Step 3

Build on a test copy

We usually build the agent against a sandbox, a safe copy of your system, on its own limited login.

Step 4

Go live with approvals

The agent saves drafts in your live system and your team approves them. Where the system keeps a change log, it shows what the agent did.

Tell us the system and the job

Name the system you run and the work you want an agent to take on. The first call is free.

Tell us the job

What an AI agent costs and how long it takes

Published ranges, with the date each was read. They describe the market, not our prices.

One automated workflow, such as invoice intake, costs $5,000 to $15,000 to build and takes two to four weeks.

Layer3 Labs, updated 2 Jul 2026, read 15 Nov 2026

Three to six connected workflows cost $15,000 to $50,000 and take six to twelve weeks.

Layer3 Labs, updated 2 Jul 2026, read 15 Nov 2026

Running costs come on top. The AI model's usage for one order of two pages is worked out at about 4 cents on our AI development services page.

Anthropic list prices, read 15 Nov 2026

Layer3 Labs sells this work itself, so read its ranges as one firm's view. Six firms' price lists, side by side, are in what an AI agent costs to build.

What we found when we checked

Notes from our own reading of vendor documents and AI answers.

Checked 9 October and 14 November 2026

SAP Business One is on none of SAP's Joule lists

Joule is SAP's AI assistant. Five SAP sources searched for “Business One” gave zero mentions, and the 236-page Joule guide of 4 November 2026 still has none.

Read the SAP Business One note
Checked 9 October and 15 November 2026

Oracle's NetSuite AI connector is free

Start there. The NetSuite page shows where it stops, in Oracle's words.

Read the NetSuite AI agents page
Checked 9 October and 15 November 2026

Odoo 20 charges credits for every AI feature

Odoo's AI service page lists “1.0 € per credit” as the price. None of 8 assistant answers we saved on 9 October mentioned the credits.

Read the Odoo AI agents page

Questions about AI agent development

What is an AI agent, in plain words?

An AI agent is a program that does one job for you inside your software. It reads something that arrives, such as an email with a purchase order attached. It checks the details against rules you set. Then it saves a record, such as a draft sales order in NetSuite, and waits for a person to approve it. Nobody has to type a prompt to start it. The email arriving starts the work.

What does an AI agent development company do?

It builds AI agents for your business. With us, the work usually runs in this order. We pick the job with you, read your system's documents and agree the rules. Then the agent gets a login of its own and is tested on your past documents. Support after launch can be part of the work too. What each build includes is agreed per project. Our team has built agents for RFQs and RFPs, bookkeeping and orders, and agents in Salesforce and HubSpot.

How much does it cost to build an AI agent?

Layer3 Labs, a US AI agency, puts one automated workflow at $5,000 to $15,000 to build, in a guide updated 2 July 2026. Other price lists reach higher. Across the six we compared, one narrow AI agent costs $5,000 to $50,000, and every list comes from a firm that sells the work. Running costs come on top: the AI model's usage and the server. All six lists are in what an AI agent costs to build.

How long does it take to build an AI agent?

Two to four weeks for one workflow, by Layer3 Labs' guide of 2 July 2026. Three to six connected workflows take six to twelve weeks. Add time to test the agent on your own past documents before it goes live. The route into your system matters too. In a second guide, updated 27 October 2026, Layer3 Labs puts a link between two modern cloud tools at 4 to 8 hours of work. A link into an older ERP through its API takes 40 to 80 hours.

Can I create my own AI agent?

Yes, for some jobs, and without code. Oracle's NetSuite AI Connector Service lets Claude or ChatGPT read and change NetSuite records, limited to your role's permissions. Oracle says the service is not a paid feature, though the AI tool may need a paid plan. Odoo 20 has an MCP server that outside AI tools can connect to. MCP is a standard way for an AI tool to use a system's functions. Hire a developer when every record must pass your own rules, or a person must approve it first.

Are AI agents worth it?

Yes, when the hours an agent saves are worth more than the build and a year of running it. Gartner expects many projects to miss that bar. In June 2025 it predicted that over 40% of agentic AI projects will be canceled by the end of 2027, because of rising costs, unclear value or weak risk controls. So count before you build. How many documents arrive a week, and how many minutes does each one take?

Will an AI agent break our ERP, or post records by itself?

Not if it can only save drafts. The usual design gives the agent a login of its own with only the rights its job needs. It saves drafts, and a person approves each one before anything posts or is sent. Many systems also keep a log of who changed what. NetSuite keeps system notes on records that no user, script or app can edit. On 14 November 2026 we asked ChatGPT, Claude, Gemini and Perplexity if it is safe to connect an AI agent to an ERP. All four said to start with read-only access.

Have you built a system with more than one AI agent?

Yes. Our team built a system of several AI agents that runs RFQs and RFPs end to end, on both sides. A customer's request becomes a draft quote, and requests for quotes go out to suppliers. An RFQ is a request for a quote. An RFP is a request for a proposal, which asks for a plan and a price. We do not name the client or give numbers from that project. The worked example on this page shows how that kind of work splits between agents.

Can you build an AI agent in Salesforce or HubSpot?

Yes. Our team has built AI agents in both CRMs. In Salesforce, an agent can sign in as an integration user. Salesforce recommends one such user for every integration, and its Minimum Access profile limits that user to the API. In HubSpot, the remote MCP server lets an agent read, write and update records such as contacts, companies, deals and tickets, as a user would. We read both vendors' pages on 16 November 2026.

Which systems do you build AI agents in?

ERPs and CRMs. Our team has taken ERPs live on SAP, Oracle NetSuite, Odoo, Microsoft Dynamics, ERPNext, Xero, QuickBooks, Sage and custom ERPs. We have built AI agents for RFQs, bookkeeping and orders in ERPs, and agents in Salesforce and HubSpot. Each system opens a different door for outside software, so we plan the route from the vendor's own documents.

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

Agentic AI is the wider term. It means AI that works through a task in steps, using tools. An AI agent is one program built that way, most often for one job. The agents on this page usually do one job in one system, with a person who approves the result. If your job has many steps that change from case to case, see agentic AI development. Four vendors' definitions sit side by side in agentic AI vs AI agents.

How is an AI agent different from a chatbot?

A chatbot waits for someone to type. An agent starts when the work arrives, such as an email with an order attached. A chatbot gives an answer on screen. An agent acts in another system: it saves a record, such as a draft quote in your ERP, for a person to approve. So an agent fits jobs that repeat many times a day.

Who owns the agent you build?

The contract for each project sets it.

What do you need from us to start?

Tell us the system you run and the job you want done. For the build itself we need the system's version, a login to a test copy, and a set of past documents with what your team entered for each one. Those past documents can become the test. The agent's drafts are checked against your team's real entries before it goes live. The first call is free.

Do you offer AI agent consulting before a build?

Yes. Before a build we check if your ERP vendor's own AI already does the job, and we plan how an agent would reach your data. Sometimes the answer is to use the vendor's AI and build nothing. That work is set out on our AI consulting services page.

Sources

Every source behind a figure or a vendor fact on this page, with the day we read it.