---
title: AI agents for Maltese firms. What is real? | Disruptive Lab
description: For companies in Malta. The large US vendors now sell AI agents. What an agent is, which tasks suit one today and what EU rules ask of you.
url: https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real
language: en-MT
---

1.  [News](https://disruptivelab.dev/en-mt/news)
2.  [AI under the company's control](https://disruptivelab.dev/en-mt/news/ai-under-control)
3.  AI agents for Maltese firms. What is real?

[Briefing](https://disruptivelab.dev/en-mt/news/briefings)  [AI under the company's control](https://disruptivelab.dev/en-mt/news/ai-under-control)

# AI agents for Maltese firms. What is real?

The large US vendors now sell software that acts across your systems on its own, and the real decision is which tasks may run without a person, and under whose control.

- **Published:** 17 November 2026
- **Updated:** 17 November 2026
- **Reading time:** 5 min read
- **Editor:** David Lahuta

## In short

-   An agent is AI software that carries out a task in several steps across your systems by itself, instead of answering one question.[6](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-6)
-   In September and October 2026, Microsoft, Salesforce, OpenAI and ServiceNow each announced new agent products, several of them still in pilot or limited availability.[1](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-1)[2](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-2)[3](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-3)[4](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-4)
-   The real decision is which tasks may run without a person, and who controls the permissions, records and approvals.
-   In Europe, the AI Act, the GDPR and where data sits come with every agent that acts.

## What did the large US vendors announce this autumn?

Four announcements, each as the vendor describes it.

Announcement

What it does

What it means for you

Act now or wait

Microsoft Copilot harness, 28 October 2026[1](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-1)

Through Microsoft Foundry, developers and enterprise customers get the layer Microsoft uses for its own Copilot products, with more than 20 AI models to choose from[1](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-1)

A building block for your developers or supplier, not a finished agent

Wait, unless your developers already build on it

Salesforce Agentforce, Dreamforce 2026[2](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-2)

Agentforce Coworker plans a task and carries it out in the CRM, and long-running agents, in pilot until November 2026, ask for approval within limits the customer sets[2](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-2)

Useful where sales and service already run in that CRM

Try one process there, otherwise wait

OpenAI Agents API, 10 September 2026[3](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-3)

A managed service for building agents that run in OpenAI's cloud, with long sessions and the use of tools[3](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-3)

A tool for building your own agents, run on the provider's side

Wait, unless you build agents yourself

ServiceNow EmployeeWorks, 15 October 2026[4](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-4)

Employees build their own AI assistants without programming, within limits set by administrators, in controlled availability[4](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-4)

Agents made by staff need company rules first

Wait, and set the rules

Microsoft says governance is built in from day one, but its announcement does not describe the controls.[1](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-1) That is the question to ask every vendor.

Menlo Ventures, a US investor, puts enterprise spending on using AI at USD 176.8 billion by September 2026, 5.4 times the year before.[5](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-5) Its survey of nearly 400 US enterprise decision makers ran in September 2026.[5](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-5) Yet the share of enterprises using AI rose only from 55% to 65%.[5](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-5) On OpenRouter, a service that passes requests on to AI models, agents now account for 80% of the text processed, though its users lean towards developers and startups.[5](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-5) Spending grows much faster than use.

## What is an agent, in plain words?

OpenAI's business guide calls agents "systems that independently accomplish tasks on your behalf".[6](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-6) A chatbot that answers one question at a time is not one.[6](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-6) It decides its next step, uses tools to read from and act in other systems, and hands control back to a person when it fails.[6](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-6)

Anthropic draws a similar line.[7](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-7) In a workflow the steps are fixed in advance, in an agent the AI model decides its own steps and tools.[7](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-7) It is the difference between a process you designed and one the software designs as it goes.

## Which tasks suit an agent today?

Anthropic advises the simplest solution that works, adding complexity only when needed.[7](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-7) OpenAI names three kinds of work where agents add value, namely judgement such as approving a refund, rules too many to maintain such as supplier security reviews, and information in documents such as an insurance claim.[6](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-6) Where fixed rules do the job, ordinary automation is enough.[6](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-6) Agents also cost more, and errors can add up from step to step.[7](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-7)

A task suits an agent when most of these hold, by both guides.[6](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-6)[7](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-7)

-   The task needs judgement, not only a fixed rule.
-   The input is e-mails, documents or free text.
-   A wrong step can be noticed and put right.
-   A person can check the result before it counts.

## What stays in the company's hands?

OpenAI's guide says sensitive, irreversible or high-stakes actions, such as cancelling orders, approving large refunds or making payments, should go to a person until trust in the agent has grown, and advises starting small with real users.[6](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-6) Anthropic recommends testing in a closed environment with limits, and letting the agent pause for a person at agreed points.[7](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-7)

The vendors build this in. Salesforce's long-running agents ask for approval,[2](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-2) and ServiceNow's assistants work within limits set by administrators.[4](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-4) Whoever holds the permissions, the records and the approvals holds the agent. Five questions are worth asking before any agent goes live.

-   Whose permissions does the agent use?
-   Where is every step recorded, and for how long?
-   Which actions wait for a person's approval?
-   Where do the data and the AI model run?
-   Can we leave the vendor and keep our data?

## What this means for a company in Europe

The AI Act binds the company that uses an agent, a deployer in the law's words, not only the one that makes it.

-   **Staff training.** Providers and deployers must take measures to support the AI literacy of their staff and of others who operate AI for them, without having to guarantee a set level.[9](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-9)[14](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-14) This applies since 2 February 2025, enforced from 2 August 2026.[8](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-8)
-   **Telling people.** From 2 August 2026, an AI system that deals directly with people must be designed so that they know they are dealing with AI.[8](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-8)[11](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-11)[15](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-15)
-   **High-risk uses.** For the uses the AI Act lists as high risk in its Annex III, the rules apply from 2 December 2027, as amended by the Digital Omnibus on AI of July 2026.[14](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-14)[8](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-8) The deployer then names people with the competence, training and authority to oversee the system, monitors how it runs and keeps its logs for at least six months.[10](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-10)
-   **Decisions about people.** Under the GDPR, a person has the right not to be subject to a decision made solely by automated means when it has legal or similarly significant effects, with some exceptions.[12](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-12) Where such a decision is allowed, the person may contest it and ask for a person to review it, as in the Commission's example of a loan decided by an algorithm.[12](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-12)
-   **Where data sits.** Personal data that an agent sends outside the European Economic Area needs safeguards so that its protection travels with it, such as an adequacy decision or standard contractual clauses.[13](https://disruptivelab.dev/en-mt/news/ai-agents-what-is-real#source-13)

An agent that acts therefore needs a named owner, records and a person who approves what matters.

## Next step

If you would like to find the first two processes in your company where an agent would save real time and stay under your control, we are glad to look at them with you in a short meeting. What you do with the result is entirely your decision.

[AI Business Harness](https://disruptivelab.dev/en-mt/ai-business-harness)

## Sources

1.  1. [Microsoft, Introducing the Microsoft Copilot harness (28 October 2026)](https://blogs.microsoft.com/blog/2026/10/28/introducing-the-microsoft-copilot-harness-unlocking-the-potential-of-truly-multi-model-agentic-work/). Read on 17 November 2026
2.  2. [Salesforce, 21 Things We Announced at Dreamforce 2026 (28 September 2026)](https://www.salesforce.com/blog/dreamforce-2026-announcements/). Read on 17 November 2026
3.  3. [OpenAI, Introducing the Agents API (10 September 2026)](https://openai.com/index/introducing-the-agents-api/). Read on 17 November 2026
4.  4. [ServiceNow, press release on EmployeeWorks (15 October 2026)](https://newsroom.servicenow.com/press-releases/details/2026/ServiceNow-introduces-the-third-generation-of-its-employee-centric-AI-harness-so-the-workforce-can-build-and-use-the-AI-their-jobs-demand/default.aspx). Read on 17 November 2026
5.  5. [Menlo Ventures, The State of Enterprise AI 2026 (3 November 2026)](https://menlovc.com/perspective/2026-the-state-of-enterprise-ai/). Read on 17 November 2026
6.  6. [OpenAI, A practical guide to building agents (2025)](https://cdn.openai.com/business-guides-and-resources/a-practical-guide-to-building-agents.pdf). Read on 17 November 2026
7.  7. [Anthropic, Building effective agents (2024)](https://www.anthropic.com/research/building-effective-agents). Read on 17 November 2026
8.  8. [European Commission, AI Act Service Desk, timeline for the implementation of the AI Act](https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act). Read on 17 November 2026
9.  9. [European Commission, AI Act Service Desk, Article 4, AI literacy](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-4). Read on 17 November 2026
10.  10. [European Commission, AI Act Service Desk, Article 26, obligations of deployers of high-risk AI systems](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-26). Read on 17 November 2026
11.  11. [European Commission, AI Act Service Desk, Article 50, transparency obligations](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50). Read on 17 November 2026
12.  12. [European Commission, Can I be subject to automated individual decision-making, including profiling?](https://commission.europa.eu/law/law-topic/data-protection/reform/rights-citizens/my-rights/can-i-be-subject-automated-individual-decision-making-including-profiling_en). Read on 17 November 2026
13.  13. [European Commission, Rules on international data transfers](https://commission.europa.eu/law/law-topic/data-protection/international-dimension-data-protection/rules-international-data-transfers_en). Read on 17 November 2026
14.  14. [Regulation (EU) 2026/1744, Digital Omnibus on AI (Official Journal of the EU, 24 July 2026)](https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng). Read on 17 November 2026
15.  15. [Regulation (EU) 2024/1689, the AI Act, consolidated text as at 27 July 2026 (EUR-Lex)](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:02024R1689-20260727). Read on 17 November 2026

## Change log

1.  17 November 2026 Published.

## Related articles

Guide · AI under the company's control

### [Shadow AI in Malta and how to control it](https://disruptivelab.dev/en-mt/news/ai-under-company-control)

For companies in Malta. Most people who use AI at work use tools the company never chose. What can leave, why bans fail and what works.

17 November 2026

Decision guide · AI under the company's control

### [Local AI or cloud AI in Malta](https://disruptivelab.dev/en-mt/news/local-or-cloud-ai)

For companies in Malta. Own machines, a European cloud or a US service, what each costs, what the rules say and who supervises AI.

17 November 2026

## Company details

- **Company name:** Disruptive Lab s.r.o.
- **Registered office:** Vavrečkova 5262, 760 01 Zlín
- **Company ID (IČO):** 08240647
- **VAT ID (DIČ):** CZ08240647
- **E-mail:** office@disruptivelab.dev
- **Phone:** +420 777 702 032
