# Agentic AI for real estate: what it is and what it does for agents

> Agentic AI for real estate plans and does the work, then waits for the agent's approval. See how it differs from chatbots and CRM automation, with five workflows.

URL: https://www.formule.io/uk/blog/agentic-ai-real-estate
Language: en
Publisher: Formule (Formule Labs Limited)
Updated: 2026-10-21
Sources checked: 2026-10-21

**Agentic AI for real estate is software that takes a goal, such as answering a new enquiry or preparing a listing, and carries out the steps itself using your agency's data and tools. It plans, acts, keeps context and stops for a person's approval before anything reaches a client. The agent decides what goes out.**

This page owns that concept. It explains how agentic AI differs from a chatbot and from CRM automation, what it does end to end on five estate agent workflows, and how to evaluate a tool before you give it access to client data. For the wider map of AI tasks, see [AI for estate agents](https://www.formule.io/uk/blog/ai-for-estate-agents).

## Agentic AI means the software does the work, not only the talking

The word "agent" has been stretched by marketing, so it helps to anchor it in the vendors' own documentation. Anthropic's engineering guide "Building effective agents" separates two designs. Workflows are "systems where LLMs and tools are orchestrated through predefined code paths", and agents are "systems where LLMs dynamically direct their own processes and tool usage". The same guide says agents can pause for human feedback at checkpoints or when they hit a blocker, and that it is common to set stopping conditions to keep control.

OpenAI's agents guide describes agents in similar terms: they can plan and complete tasks using tools, and they maintain context across steps. Two ideas run through both definitions. The software chooses its own steps toward a goal, and it uses tools to act, not only to write text. For an estate agent that translates into a simple test: after you give it a goal, does it go and do the work in your systems, or does it hand you text to paste somewhere?

A second test matters just as much in a regulated, relationship-driven business. Does it stop for you? Good agentic design keeps a person at the points where a mistake costs money or trust: sending a message, publishing a listing, promising a viewing. Agentic does not mean autonomous. It means the AI does the preparation, and you make the decision.

## A chatbot, CRM automation and an AI agent differ in who starts the work

These three get confused because all of them can touch an enquiry. The table separates them by what they actually do.

| Dimension | Chatbot | CRM automation | AI agent |

| Who starts the work | The visitor or you, by asking a question | A trigger you configured, such as a new lead or a date | You give a goal, or an event arrives, and the agent decides the steps |

| Memory | Usually the current conversation only | Fields and statuses in the record, no understanding of context | Your records plus the conversation history, used to shape each task |

| Actions | Answers in text | Fixed actions: send a template, change a status, create a task | Chooses and uses tools: reads the file, drafts, builds a page, updates records |

| Approval | Often none, because it replies directly to the visitor | None after setup, the rule fires every time | Built in by design: you approve what leaves the building |

| Typical examples | A website widget that answers opening hours | A drip email sent three days after an enquiry | An assistant that reads an enquiry, checks the property, drafts the reply and proposes a viewing slot |

None of the three replaces the others. Automation is still the right tool for fixed, repeatable steps such as reminders. A chatbot is fine for simple public questions. An agent earns its place on work that varies each time and needs your data, like a reply that has to match one specific buyer and one specific property.

## What an agent does end to end on five real estate workflows

The five workflows below follow the same pattern: the AI does the preparation, the agent approves, and clear limits stay with the human. The examples are illustrative and describe what agentic AI can do in general; the list in the Really section below shows what Really does today.

### New enquiry

**What the AI does:** reads the enquiry in the buyer's language, finds the property it refers to, checks its status and your diary, and drafts a short personal reply with two real viewing slots and one question about the buyer's timing.

**What the agent approves:** the wording, the slots and the send. Nothing goes to the buyer until you say so.

**Limits:** the draft is only as accurate as the record. If the property was withdrawn yesterday and the status was not updated, the draft will be wrong with confidence. It must never invent a price reduction or confirm a slot you have not offered. The full set-up is in [AI lead follow-up on WhatsApp](https://www.formule.io/uk/blog/ai-lead-follow-up-whatsapp).

### Listing from a link

**What the AI does:** takes a property link or a set of facts and photos and produces a structured description, key features in a sensible order, a shareable listing page and a short social caption.

**What the agent approves:** every fact against the file, such as room sizes, tenure, energy rating and what is included, before the page is shared.

**Limits:** a model reads a sunny photo and writes "south-facing garden". Facts that are not in the source should be marked as missing, not guessed. Steps and a checklist are in [turning a property link into a listing](https://www.formule.io/uk/blog/ai-property-listing-from-a-link).

### Weekly pipeline review

**What the AI does:** reads your pipeline and lists the enquiries with no reply for three days, vendors who have not heard from you this month and buyers who match a new instruction, each with a suggested next step.

**What the agent approves:** which follow-ups to act on and the wording of each message. The review informs your Monday plan; it does not run it for you.

**Limits:** an agent that can read your pipeline can also reveal it. Keep access read-only where possible and use it only inside tools you have vetted. Connecting an assistant to your CRM data safely is covered in [ChatGPT and Claude for estate agents](https://www.formule.io/uk/blog/chatgpt-claude-for-estate-agents).

### Marketing assets

**What the AI does:** turns an approved listing into social copy, images and a short video, in the formats each channel needs.

**What the agent approves:** every image and clip before publication, and the label that tells buyers what is staged or generated.

**Limits:** AI must not change structure, size, condition or surroundings in a way a buyer would rely on. Disclosure rules vary by country; the guide to [AI virtual staging and property video](https://www.formule.io/uk/blog/ai-virtual-staging-property-video-rules) covers the UK, Germany, France, Spain, Italy and Poland.

### Compliance check

**What the AI does:** prepares the paperwork around compliance. It compares your privacy notice with the tools you use, drafts a checklist for a new tool and flags places where client data leaves your control.

**What the agent approves:** the final policy, the lawful basis for each use of personal data and any decision to adopt the tool.

**Limits:** AI is not your compliance officer or your solicitor. It can miss a local rule, and its summary of a law is not legal advice. Use it to prepare questions and drafts, then confirm with the right professional.

## Evaluating an agentic AI tool for your agency starts with the approval model

Marketing pages for agentic tools look alike. These six checks separate a tool you can trust with client work from one that only demos well. Run each one with the vendor in writing, and test it yourself with a live enquiry.

- **Approval model.** Find out exactly which actions need your approval, and whether you can change that. Messages to clients, published listings and changes to records should all wait for you by default. Be wary of any tool that describes itself as fully autonomous.

- **Data access.** List what the agent can read and what it can change: contacts, conversations, documents, diary. Prefer read access first, narrow permissions, and a clear answer on where data is stored and whether it is used to train models.

- **Audit trail.** You should be able to see what the agent did, when, on whose approval and from which data. If something goes wrong with a client, you need that record.

- **Languages.** Test the tool in the languages your buyers actually write in, with real examples. A reply in a buyer's own language that is stiff or wrong costs you the lead.

- **Channels.** Check which channels it works on: WhatsApp, email, portal enquiries, voice. An agent that only lives in a separate window adds a step instead of removing one.

- **EU AI Act and GDPR.** Ask how the vendor handles personal data under GDPR and how it meets the EU AI Act transparency rules. Article 50 splits its duties between providers (the companies that build the AI system) and deployers (the businesses that use it, which includes an estate agency). Providers must design systems that interact directly with people so those people are informed they are dealing with an AI system, unless that is obvious from the context, and must mark synthetic audio, image, video or text outputs in a machine-readable format, with exemptions for assistive editing and for systems that do not substantially alter the input. Deployers have their own duty: if you publish image, audio or video content that is a deep fake, you must disclose that it was artificially generated or manipulated. Ask the vendor what it does as provider, and check what you owe as deployer. The practical check is in [AI, GDPR and the EU AI Act for estate agents](https://www.formule.io/uk/blog/ai-gdpr-eu-ai-act-estate-agents).

The EU AI Act text is published by the European Commission's AI Act Service Desk, and the consolidated version it shows was current as of 27 July 2026. Rules and dates change, so check the current text before you rely on a deadline, and ask your regulator or adviser about duties specific to your market.

## Really is Formule's agentic AI for estate agents

Formule is a CRM with agentic AI. The CRM keeps the record, and Really, Formule's AI agent, does the work around it. You talk to Really in the Formule app, on [WhatsApp](https://www.formule.io/whatsapp?lang=en) or by voice, in English, German, French, Spanish, Italian or Polish. These are the things Really does today:

- **Set-up in one chat.** New agents are set up by Really in a single conversation.

- **Lead follow-up.** Really helps you follow up leads and drafts the replies. You approve what goes out.

- **Listings from a link.** Paste a property link and Really turns it into a listing draft and a shareable listing page.

- **Marketing assets.** The Marketing section creates AI images and short videos of a property, paid for with credits.

- **One pipeline.** Leads from ads and portals arrive in a single pipeline, and branch managers see every lead.

- **Your own assistant.** Formule connects to Claude and ChatGPT through its MCP server, so you can ask questions about your pipeline from those tools.

Really drafts and acts on the agent's approval. It does not send on its own, and it is not fully autonomous. That is a design choice, not a missing feature: in a business built on personal relationships, the agent decides what a client reads.

## A sensible first step is one workflow with approval switched on

Pick the workflow that costs you the most hours, usually replying to enquiries or writing listings. Run it on five real cases, edit every draft and note each correction. Those corrections become your rules for the agent. Then decide whether to add a second workflow. The week-long plan in [AI for estate agents](https://www.formule.io/uk/blog/ai-for-estate-agents) shows the sequence day by day.

To try it with Really, [start free](https://app.formule.io/signup?lang=en); the free start includes a 7-day Really trial. You can also see how agents use Really day to day on the [Formule page for your market](https://www.formule.io/uk).

## The five guides cover each task in detail

- [AI lead follow-up on WhatsApp](https://www.formule.io/uk/blog/ai-lead-follow-up-whatsapp) for fast, personal replies you approve.

- [A listing from a property link](https://www.formule.io/uk/blog/ai-property-listing-from-a-link) for descriptions, pages and social copy.

- [AI photos, staging and video](https://www.formule.io/uk/blog/ai-virtual-staging-property-video-rules) and the disclosure rules per market.

- [ChatGPT and Claude with your CRM](https://www.formule.io/uk/blog/chatgpt-claude-for-estate-agents) for questions about your own pipeline.

- [GDPR and the EU AI Act](https://www.formule.io/uk/blog/ai-gdpr-eu-ai-act-estate-agents) for a practical check before using client data.

Sources: [Anthropic, Building effective agents](https://www.anthropic.com/engineering/building-effective-agents); [OpenAI, Agents guide](https://developers.openai.com/api/docs/guides/agents); [European Commission AI Act Service Desk, Article 50](https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50).

## Sources
- https://www.anthropic.com/engineering/building-effective-agents
- https://developers.openai.com/api/docs/guides/agents
- https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50