import { classify, obligations } from '@eu/ai-act'; // Regolamento (UE) 2024/1689import { classify, obligations } from '@eu/ai-act'; // Regolamento (UE) 2024/1689import { classify, obligations } from '@eu/ai-act'; // Regolamento (UE) 2024/1689import { classify, obligations } from '@eu/ai-act'; // Regolamento (UE) 2024/1689import { classify, obligations } from '@eu/ai-act'; // Regolamento (UE) 2024/1689import { classify, obligations } from '@eu/ai-act'; // Regolamento (UE) 2024/1689

const system = inventory.find((s) => s.name === 'chatbot-assistenza'); // art. 50: trasparenzaconst system = inventory.find((s) => s.name === 'chatbot-assistenza'); // art. 50: trasparenzaconst system = inventory.find((s) => s.name === 'chatbot-assistenza'); // art. 50: trasparenzaconst system = inventory.find((s) => s.name === 'chatbot-assistenza'); // art. 50: trasparenzaconst system = inventory.find((s) => s.name === 'chatbot-assistenza'); // art. 50: trasparenzaconst system = inventory.find((s) => s.name === 'chatbot-assistenza'); // art. 50: trasparenza

const role = system.provider === company ? 'fornitore' : 'utilizzatore'; // art. 3const role = system.provider === company ? 'fornitore' : 'utilizzatore'; // art. 3const role = system.provider === company ? 'fornitore' : 'utilizzatore'; // art. 3const role = system.provider === company ? 'fornitore' : 'utilizzatore'; // art. 3const role = system.provider === company ? 'fornitore' : 'utilizzatore'; // art. 3const role = system.provider === company ? 'fornitore' : 'utilizzatore'; // art. 3

if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');

if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });

await notice.show('Stai parlando con un assistente virtuale basato su AI');await notice.show('Stai parlando con un assistente virtuale basato su AI');await notice.show('Stai parlando con un assistente virtuale basato su AI');await notice.show('Stai parlando con un assistente virtuale basato su AI');await notice.show('Stai parlando con un assistente virtuale basato su AI');await notice.show('Stai parlando con un assistente virtuale basato su AI');

await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });

const decision = await human.review(cv); // D.Lgs. 179/2026, art. 41const decision = await human.review(cv); // D.Lgs. 179/2026, art. 41const decision = await human.review(cv); // D.Lgs. 179/2026, art. 41const decision = await human.review(cv); // D.Lgs. 179/2026, art. 41const decision = await human.review(cv); // D.Lgs. 179/2026, art. 41const decision = await human.review(cv); // D.Lgs. 179/2026, art. 41

await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });

const fine = Math.min(35_000_000, revenue * 0.07); // PMI: il più basso, art. 99const fine = Math.min(35_000_000, revenue * 0.07); // PMI: il più basso, art. 99const fine = Math.min(35_000_000, revenue * 0.07); // PMI: il più basso, art. 99const fine = Math.min(35_000_000, revenue * 0.07); // PMI: il più basso, art. 99const fine = Math.min(35_000_000, revenue * 0.07); // PMI: il più basso, art. 99const fine = Math.min(35_000_000, revenue * 0.07); // PMI: il più basso, art. 99

await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42

await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13

export default compliance.by({ design: true, deadline: '2027-12-02' });export default compliance.by({ design: true, deadline: '2027-12-02' });export default compliance.by({ design: true, deadline: '2027-12-02' });export default compliance.by({ design: true, deadline: '2027-12-02' });export default compliance.by({ design: true, deadline: '2027-12-02' });export default compliance.by({ design: true, deadline: '2027-12-02' });

import { classify, obligations } from '@eu/ai-act';                         // Regolamento (UE) 2024/1689
const system = inventory.find((s) => s.name === 'chatbot-assistenza');        // art. 50: trasparenza
const role = system.provider === company ? 'fornitore' : 'utilizzatore';      // art. 3
const risk = classify(system, { annexIII: true, profiling: false });          // art. 6
if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');
if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });
await notice.show('Stai parlando con un assistente virtuale basato su AI');
await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });
const decision = await human.review(cv);                                      // D.Lgs. 179/2026, art. 41
await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });
const fine = Math.min(35_000_000, revenue * 0.07);                            // PMI: il più basso, art. 99
await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42
await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13
export default compliance.by({ design: true, deadline: '2027-12-02' });
import { classify, obligations } from '@eu/ai-act';                         // Regolamento (UE) 2024/1689
const system = inventory.find((s) => s.name === 'chatbot-assistenza');        // art. 50: trasparenza
const role = system.provider === company ? 'fornitore' : 'utilizzatore';      // art. 3
const risk = classify(system, { annexIII: true, profiling: false });          // art. 6
if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');
if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });
await notice.show('Stai parlando con un assistente virtuale basato su AI');
await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });
const decision = await human.review(cv);                                      // D.Lgs. 179/2026, art. 41
await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });
const fine = Math.min(35_000_000, revenue * 0.07);                            // PMI: il più basso, art. 99
await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42
await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13
export default compliance.by({ design: true, deadline: '2027-12-02' });
import { classify, obligations } from '@eu/ai-act';                         // Regolamento (UE) 2024/1689
const system = inventory.find((s) => s.name === 'chatbot-assistenza');        // art. 50: trasparenza
const role = system.provider === company ? 'fornitore' : 'utilizzatore';      // art. 3
const risk = classify(system, { annexIII: true, profiling: false });          // art. 6
if (risk === 'vietato') throw new Error('Pratica vietata dal 2 febbraio 2025');
if (risk === 'alto') await obligations.highRisk(role, { from: '2027-12-02' });
await notice.show('Stai parlando con un assistente virtuale basato su AI');
await training.log({ team: 'marketing', topic: 'alfabetizzazione AI', art: 4 });
const decision = await human.review(cv);                                      // D.Lgs. 179/2026, art. 41
await workers.inform({ tool: system.name, legge: '132/2025', art: 11 });
const fine = Math.min(35_000_000, revenue * 0.07);                            // PMI: il più basso, art. 99
await dvr.add({ system: system.name, impatto: 'organizzazione del lavoro' }); // art. 42
await client.inform({ studio: true, strumenti: ['assistente di ricerca'] }); // art. 13
export default compliance.by({ design: true, deadline: '2027-12-02' });

AI Agents

AI agents for business:
what they are, examples and costs

An AI agent is software that takes a goal and completes it in several steps: it decides what to do, uses tools such as email, your CRM or your ERP, checks the result and asks a person to approve sensitive actions. A chatbot answers; an agent acts.

This guide covers examples by team, list prices read on vendor sites on 10 November 2026, the EU AI Act and GDPR rules that apply, and 6 steps to start without wasting budget.

By Cesare D'Adamo/Updated 10 November 2026/10 min read

Book a free consultation All guides

In this guide

01What is an AI agent, and how is it different from a chatbot?
02How many companies use AI agents today?
03AI agent examples by team
04What is new in 2026: agents inside the tools you already use
05How much does an AI agent cost?
06Which rules apply to AI agents: EU AI Act and GDPR
07How to implement an AI agent in your company: 6 steps
08Frequently asked questions

What is an AI agent, and how is it different from a chatbot?

An AI agent is a system built on a language model that pursues a goal by using tools, instead of only replying. OpenAI describes agents as systems that complete tasks independently on your behalf, built from three parts: the model that reasons, the tools it can call (APIs, databases, applications) and the instructions that set its limits.

Anthropic's "Building effective agents" (19 December 2024) draws a useful line: in a workflow the model follows paths written in code; in an agent the model decides the steps and which tools to use. The European Commission notes that "AI agent" has no legal definition: it is a system that takes input from its environment and acts on it, for example by calling functions.

Aspect	Chatbot	Classic automation	AI agent
What it does	Answers questions	Runs fixed rules: if A, then B	Pursues a goal over several steps
Who decides the steps	Nobody	Whoever designed the flow	The model, within the limits you set
Main risk	A wrong answer	Breaks on a case nobody planned for	A wrong action in a live system

01GoalExample: handle order status requests from customers
02PlanThe model splits the task into steps and picks its tools
03ActReads the order system, replies to the customer, updates the CRM
04CheckVerifies the result of each step before moving on
05Human approvalRefunds, payments and cancellations go to a person

How an AI agent works. Simplified from OpenAI's and Anthropic's guidance on building agents.

How many companies use AI agents today?

Mostly large companies, so far. McKinsey's survey "The state of AI in 2026" (August 2026, 1,719 respondents in 97 countries, fielded May to June 2026) defines agents as AI systems that plan and execute multiple workflow steps on their own, and asks who has moved beyond pilots.

Large organizations, 202640%
Large organizations, 202527%
Smaller organizations, 202622%

Respondents scaling AI agents in at least one function. Large means over $1 billion in revenue. Source: McKinsey, The state of AI in 2026, August 2026.

The gap is widening: large organizations grew from 27% to 40% in a year, while smaller ones stayed at 22%. Agents are scaled most often in IT, knowledge management and software engineering. Value is still concentrated: only about 6% of respondents qualify as AI high performers, and 37% attribute at least some EBIT impact to AI.

AI agent examples by team

A good first agent works on a frequent process where someone copies data between systems today. OpenAI suggests agents for three situations: decisions with many exceptions, rule sets that have become costly to maintain, and unstructured data such as emails and documents.

Team	What the agent can do	Where a person decides
Customer service	Reads the request, checks the order, replies, opens the return	Refunds above a threshold, complaints
Sales	Qualifies inbound leads, updates the CRM, drafts the follow up	Quotes and discounts
Finance	Matches supplier invoices to orders and delivery notes, flags differences	Payments and final postings
Procurement	Compares quotes, prepares the periodic supplier review	Choosing the supplier and signing
IT and internal helpdesk	Routes tickets, answers procedure questions, collects details for the technician	Access rights and system changes
HR	Answers questions on leave, policies and internal procedures	Hiring and performance decisions

The last row is a hard line in the EU. An agent that filters applications, evaluates candidates or monitors worker performance falls under the uses listed in Annex III of the AI Act, which makes it high-risk with much heavier duties. For a first project, stay with internal, repetitive and reversible work.

What is new in 2026: agents inside the tools you already use

Often you no longer build an agent from scratch: the major vendors are putting agents into workplace software, with admin controls.

ChatGPT Business includes workspace agents, which run on a schedule or on request, use approved connectors such as a CRM, ticketing or Gmail, and can stop for approval before sensitive actions.
Gemini Enterprise Business includes Workflow Builder, a no-code agent designer for multi-step tasks across several apps.
Microsoft introduced Autopilot on 25 September 2026, a cloud-hosted digital teammate with its own identity in your tenant, in private preview with usage-based billing. Agent 365, its control plane to inventory, govern and secure agents, has been generally available since 1 May 2026.
Anthropic introduced Claude Duet on 21 October 2026, a proactive agent that runs on its own virtual machine and browser and asks before it sends messages, pays, signs or deletes. It is a research preview with a waitlist that started in the US.

Agents that browse on someone's behalf also read your website, one more reason to invest in generative engine optimization.

Underneath sits a shared standard. The Model Context Protocol (MCP), created at Anthropic, connects agents to company tools and data in one consistent way. Since 9 December 2025 it is hosted by the Linux Foundation's Agentic AI Foundation, which counted more than 10,000 published MCP servers at launch. For you that means fewer custom integrations and less vendor lock-in.

How much does an AI agent cost?

You pay for an agent in one of three ways: per user per month when it comes with a suite, per use (conversations, actions or credits) on dedicated platforms, or per token if you build on an API. These are list prices on vendor sites, read on 10 November 2026, before any project costs.

Offer	Pricing model	List price
ChatGPT Business	Per user per month	$20 billed annually, $25 billed monthly
Gemini Enterprise Business	Per user per month	From $21, with a 30-day free trial
Microsoft 365 Copilot	Per user per month	$30 paid yearly; internal agents included, with some exceptions
Microsoft Copilot Studio	Credit pack	$200 per month for 25,000 Copilot Credits, or pay as you go
Salesforce Agentforce	Per conversation or per action	$2 per conversation, or $500 per 100,000 Flex Credits (20 credits per action)
Claude API, for a custom agent	Per million tokens	Haiku 5.5: $0.10 input and $0.50 output; Sonnet 5.5: $2 and $10

Run the numbers on your volume. For example, 1,000 customer requests a month on Agentforce cost $2,000 at the per-conversation price; if each request needs 5 actions, Flex Credits come to 100,000 credits, or $500. The right model depends on how many actions it takes to close a case.

The list price is only part of it: Anthropic points out that agentic systems trade latency and cost for better results. Budget for connecting systems, cleaning data, testing and weekly review, the items that decide whether the project pays back; estimating them for your process is a good use of a free first consultation.

Which rules apply to AI agents: EU AI Act and GDPR

AI agents have no rulebook of their own: the rules follow the system and what you use it for. In its AI Act Service Desk FAQ, the European Commission says agents are not a separate category, fall under the definition of an AI system, and that its views on them are still preliminary.

Rule	When it applies to your agent	From
AI Act, Article 4	Always: measures for the AI literacy of people who use or run it	2 February 2025
AI Act, Article 50	It interacts with people or generates content	2 August 2026
AI Act, Annex III	Hiring, worker management, credit scoring, life and health insurance pricing and other listed uses	2 December 2027
GDPR, Article 22	Decisions based solely on automated processing with legal or similarly significant effects	Already in force

Under GDPR Article 22, people have the right not to be subject to decisions based solely on automated processing with legal or similarly significant effects. Even under the exceptions (contract, law or explicit consent), they can obtain human intervention and contest the decision. The prudent setup: the agent prepares a credit decision, a person closes it.

National rules can go further. Italy's Legislative Decree 179/2026 (Official Journal, 8 October 2026), for example, reserves final decisions on hiring, evaluations, discipline and dismissal to a person with real authority, and makes purely automated ones void. If your agent talks to customers, check the AI Act transparency timeline for Article 50.

How to implement an AI agent in your company: 6 steps

Start from a process, not a tool.

01Pick one process that is frequent and measurable. Write down today's volume, handling time and error rate: that is your baseline.
02Try the simplest solution first, such as one model call with the right documents, or a rules-based automation. Anthropic advises adding complexity only when it clearly improves results.
03Grant only the access it needs: a dedicated account, least privilege, no payments or sensitive data unless the task requires them.
04Decide where a person must approve. OpenAI names two cases: high-risk or irreversible actions, such as cancelling orders, large refunds and payments, and an agent that exceeds its retry limits.
05Test in a sandbox with real past cases, then on a small share of live volume, with caps on steps and spend per task.
06Govern it and train people: keep a register of agents with an owner, access rights and logs, update your AI use policy, and take the AI literacy measures that Article 4 of the AI Act asks for.

The typical risk comes from outside content. The OWASP Top 10 for Agentic Applications (9 December 2025) ranks agent goal hijack first, for example hidden instructions in an email or web page that push the agent to do something else, followed by tool misuse and identity and privilege abuse.

Whiteboard covered with colorful sticky notes used to organize tasks — Before the agent comes the process map: steps, exceptions, who approves what. Photo by cottonbro studio / Pexels

The companies that get most value redesign the process: in McKinsey's survey nearly three quarters of high performers had fundamentally redesigned workflows, against about a quarter of the rest. That is the work of an AI transformation program. If nobody inside can own it yet, an interim or fractional leader can set it up and hand it over.

Free consultation

Want to apply this to your company?

In a free first call we look at your situation and I tell you where I would start, with which tools and what results to expect.

Book a free consultation Services

Frequently asked questions

What is the difference between an AI agent and a chatbot?

A chatbot answers questions. An agent takes a goal, chooses the steps, uses tools such as your CRM, ERP or email, and checks the result.

How much does an AI agent cost for a small business?

As of 10 November 2026, agents bundled in suites start at $20 to $21 per user per month (ChatGPT Business, Gemini Enterprise Business). Usage-based platforms charge per conversation or action, for example $2 per conversation on Agentforce. Add integration, testing and oversight.

Do I need a developer to build an AI agent?

Not always. Gemini Enterprise's Workflow Builder, a no-code designer, and ChatGPT Business workspace agents let you build simple agents on apps you already connect. Your ERP, permissions or an API need technical skills.

What does the EU AI Act say about AI agents?

The European Commission says agents are not a separate category: the rules for AI systems apply. Transparency duties apply from 2 August 2026 if the agent interacts with people or generates content, and high-risk duties for Annex III uses from 2 December 2027.

Is it safe to connect an AI agent to email and business systems?

Yes, with a dedicated account, least privilege, action logs and approvals for sensitive operations. OWASP ranks goal hijack first, hidden instructions in outside content that redirect the agent, so test for it before going live.

What is MCP and why does it matter?

The Model Context Protocol is an open standard, created at Anthropic, for connecting agents to tools and data. Since 9 December 2025 it is hosted by the Linux Foundation's Agentic AI Foundation, which says ChatGPT, Gemini, Microsoft Copilot and Claude are among the platforms that adopted it.

Sources

Sources checked on 10 November 2026.

About the author

Cesare D'Adamo

Product and AI strategy consultant

15+ years in product, growth and AI, more than 20 AI-powered products built, and executive teams trained on EU AI Act compliance. Works from Berlin with companies across Europe, in English and Italian.

LinkedIn Full profile Services

Let's put AI to work

Strategy, products and automation with AI that deliver measurable results, within the rules. It starts with a free call.

Book a free consultation

First call free, no commitment

AI agents for business:
what they are, examples and costs

In this guide

What is an AI agent, and how is it different from a chatbot?

How many companies use AI agents today?

AI agent examples by team

What is new in 2026: agents inside the tools you already use

How much does an AI agent cost?

How to implement an AI agent in your company: 6 steps

Want to apply this to your company?

Frequently asked questions

What is the difference between an AI agent and a chatbot?

How much does an AI agent cost for a small business?

Do I need a developer to build an AI agent?

What does the EU AI Act say about AI agents?

Is it safe to connect an AI agent to email and business systems?

What is MCP and why does it matter?

Sources

Cesare D'Adamo

Related guides

GEO: how to get cited by ChatGPT, Gemini and AI Overviews

/services/ai-transformation

/services/interim-leadership

/services/growth-systems

All guides

Let's put AI to work

AI agents for business: what they are, examples and costs

In this guide

What is an AI agent, and how is it different from a chatbot?

How many companies use AI agents today?

AI agent examples by team

What is new in 2026: agents inside the tools you already use

How much does an AI agent cost?

Which rules apply to AI agents: EU AI Act and GDPR

How to implement an AI agent in your company: 6 steps

Want to apply this to your company?

Frequently asked questions

What is the difference between an AI agent and a chatbot?

How much does an AI agent cost for a small business?

Do I need a developer to build an AI agent?

What does the EU AI Act say about AI agents?

Is it safe to connect an AI agent to email and business systems?

What is MCP and why does it matter?

Sources

Cesare D'Adamo

Related guides

GEO: how to get cited by ChatGPT, Gemini and AI Overviews

/services/ai-transformation

/services/interim-leadership

/services/growth-systems

All guides

Let's put AI to work

AI agents for business:
what they are, examples and costs