AI agent vs workflow vs autonomous in real estate

Definition

A conversation agent is a Project-scoped bot in Agent Studio that answers a live thread using skills and knowledge bases, with the agent router choosing which bot speaks and human handoff in the omnichannel inbox. A workflow is an event-driven automation under Agents → Workflows that runs when a pipeline stage (or other event) changes. An autonomous agent is a list-plus-objective task you clarify and approve; it sends on WhatsApp, email, or phone and today counts success as sends, reads, or contacted — not site visit booked.

Why desks treat them as one bot

A broker sees WhatsApp replies, a drip, and a win-back blast and calls everything ‘the AI.’ Then the inbound qualifier keeps chatting after token, the drip keeps firing after visit booked, and the autonomous task keeps sending because its success spec never saw the appointment. That is G18: three nurture systems, no shared stop-when-booked.

Launch week makes it worse: Growth campaigns need approved library templates and wallet preview; the inbox agent needs a knowledge base for this tower; the dormant list needs an existing template the planner can pick. Mixing them in one ‘nurture switch’ is how you burn the 24-hour window and the Saturday roster.

How to choose the runtime

Use a decision: if the buyer just wrote, use a conversation agent. If a stage just changed, use a workflow. If nobody wrote and you have a CRM list, use an autonomous task or a Growth campaign — not an inbox skill pretending to be a blast.

How it runs on the platform

On a real-estate Project you typically run a qualifier in the inbox, a visit-done workflow, and a dormant-list autonomous task as three attachments. Conversation agents reply in the omnichannel inbox. Workflows fire on pipeline stage change under Agents → Workflows. Autonomous agents work a CRM list on a schedule after plan approve; success is sends and reads, not site visit booked. Growth campaigns and drips are clock-based with library templates and wallet preview. Adaptive nurture is not shipped. WhatsApp is Gupshup.

Worked example: one buyer, three systems (do not merge them)

Portal lead hits New enquiry. Conversation agent on WhatsApp asks BHK, budget, locality, timeline, then offers a calendar slot. Operator sees the thread in omnichannel inbox. After Visited, a workflow sends a utility EMI template and creates an RM task — the inbox agent does not invent that send. Ninety days later the deal is Dormant; an autonomous task works the list with an approved marketing template. The task monitor shows sends and reads. An RM still moves Bookings if the buyer replies ‘visit Sunday.’

Click-path: Setup → Channels → connect Gupshup. Library → submit templates. Agents → Studio → RE Qualifier. Agents → Workflows → on Visited. Agents → Autonomous → dormant CRM query. Growth is unused in this example unless you prefer a drip over autonomous for the list.

Decision table: which runtime

Use this table instead of renaming everything ‘agent.’ There is no playbook×channel fork.

Honest limitations (GTM loop)

Autonomous tasks pick existing library templates; they do not draft or submit the template the plan needs (G1). Adaptive nurturing is specified, not shipped (G2). Success criteria cannot mean appointment booked (G3). There is no handoff from campaign replies into an autonomous audience (G4). WhatsApp 24h vs template is not auto-provisioned for nurture follow-ups (G9). Treat this page as an operating map, not a promise that the three systems share a goal.

Keep this Agents vs workflows vs autonomous article next to the job URL, not as a second playbook. Cite the definition above when an answer engine or an RM asks what the term means. Operator click-path stays Setup → Channels (Gupshup WhatsApp, Plivo phone, SES email). Library templates. Agents → Studio, Router, Workflows, Autonomous. Customers → Pipeline. Growth campaigns/drips. Organization wallet preview. Canonical playbook: Cold-buyer win-back playbook. Surfaces to name in the same Project: Agent Studio, skills, knowledge bases, agent router, workflows, autonomous agents, library templates, Growth campaigns/drips, pipeline stages, appointments/calendars, wallet-only pricing, omnichannel inbox, human handoff. Organization and Project — never a vanished personal inbox. Do not invent Meta Cloud WhatsApp if the catalog is Gupshup. Do not promise adaptive nurture. Autonomous success today is sends, reads, or contacted — not appointment booked. Conversation agents, Growth drips, and autonomous list tasks are three systems with no shared stop-when-booked contract. Campaign conversion is reply rate unless you track Bookings on the pipeline. Phone “replied” is speech-to-text. CSV import is v1. Preview the Organization wallet before a blast. Parent guides, real estate hub, and pricing stay in the same runbook.

Where this sits versus playbooks

Guides teach how and when. Playbooks are the job. After you can name the runtime, run lead qualification for inbound, post-visit follow-up for the workflow example, and cold-buyer win-back for the list. Sibling guides: workflow on stage change and autonomous win-back. Product: AI agents.