AI for PropTech
AI

AI for PropTech

June 9, 2026 6 min read

Prompts vs Skills vs Workflows vs Agents

Most “AI tips” for real estate lump four different things together and call them all workflows. They are not the same, and the difference decides how much time you actually save. A prompt is one-shot. A skill is set up once and reused. A workflow chains steps across tools. An agent runs on its own toward a goal. Pick the wrong one and you either retype the same instructions forever or hand a deal to a bot that was never built to run it.

This guide defines all four in plain terms, with a CRE example for each, so you know what you are actually looking at on a tool page.


Key takeaways

  • Prompt: a single instruction you paste for one answer. Not saved, not autonomous. Example: “Extract the parties, price, and contingencies from this PSA.”
  • Skill: a reusable instruction bundle you set up once, then trigger automatically. Claude Skills and ChatGPT Custom GPTs are skills. Example: an “LOI Drafter” that fills your branded template from any offering memorandum.
  • Workflow: a defined multi-step process, often across several tools or automated. Example: a new lead is classified, logged in your CRM, and answered, with no human touch.
  • Agent: an autonomous system that plans and executes a multi-step task toward a goal. Example: Claude Code or Manus producing a full due-diligence package from a folder of deal documents.
  • Stack them: use prompts to think, skills for recurring tasks, workflows to connect systems, and agents to run the whole thing.

Prompt: one instruction, one answer

A prompt is text you send to a model for a single response. It can be one line or a long brief with your underwriting criteria attached. It is not saved as its own tool, it has no triggers, and it does nothing until you press send.

Use a prompt for ad-hoc thinking: a quick lease summary, a memo draft, a one-off market question. The catch is repetition. If you run the same instruction on every deal, you are retyping or pasting it each time, because the system does not treat your prompt as a reusable thing.

CRE example: you paste a rent roll and T-12 into ChatGPT and ask it to underwrite the deal at 65 percent LTV with a 6 percent exit cap. You get one analysis. The next deal needs the prompt again.

Skill: set it up once, reuse it forever

A skill is a reusable instruction bundle. You define how the AI should handle a recurring task once, give it a name, and it loads automatically when your request matches. Claude Skills and ChatGPT Custom GPTs are the common examples. A skill can carry files too, like your branded template or a sample output, so the result is consistent every time.

The difference from a prompt is persistence and automatic triggering. You do not restate the instructions. You drop in the document and the skill runs.

CRE example: an “LOI Drafter” skill in Claude holds your letter-of-intent template, logo, and formatting. Attach any offering memorandum, say “draft an LOI,” and it produces a branded draft and asks only for the missing terms. Build it once, use it on every deal.

Workflow: multiple steps, often across tools

A workflow is a defined sequence of steps, usually across more than one tool, often triggered by an event or a schedule. No-code platforms like n8n and Make build these, and so do multi-tool chains where the output of one tool feeds the next. A workflow has fixed logic: you design the path, and it runs that path every time.

The difference from a skill is scope. A skill is AI-side behavior for one task. A workflow is system plumbing that connects email, CRM, spreadsheets, and AI steps into one repeatable process.

CRE example: a new website inquiry triggers a workflow. The text is classified by asset type and budget, a CRM record is created, a personalized follow-up is drafted and sent, and the acquisitions channel gets a Slack ping. It runs on every lead without you touching it.

Agent: autonomous, goal-driven

An agent is a system that acts on its own toward a goal. You give it an objective, and it breaks the work into steps, decides the sequence, calls tools, and iterates until it is done, with little step-by-step prompting. Claude Code, Claude Cowork, Manus, and Perplexity’s agentic modes are agents. The key trait is autonomy: it keeps working after the kickoff.

The difference from a workflow is who designs the steps. A workflow follows a path you defined. An agent figures out the path itself.

CRE example: you point Claude Code at a folder of deal documents and ask for a due-diligence package. It reads the files, builds a two-scenario pro forma, a red-flag report, a checklist, and seller questions, then hands back six deliverables. You set the goal; it ran the steps.


Quick comparison

TypeWhat it isReuseAutonomyCRE example
PromptOne instruction, one answerManual copy-pasteNone”Summarize this lease.”
SkillReusable instruction bundle, set up onceSaved, auto-triggeredNone on its own”LOI Drafter” that fills your template
WorkflowMulti-step process across toolsRuns on every triggerLimited to its designLead to CRM to follow-up, automated
AgentAutonomous, goal-driven systemReusable for many goalsHighClaude Code building a DD package

How to stack them

These are layers, not rivals. Use prompts for one-off thinking. Wrap the patterns you repeat into skills. Wire your systems together with workflows. Put an agent on top when you want a goal handled end to end. A strong CRE stack uses all four: a skill abstracts the lease, a workflow routes it, and an agent keeps the deal moving while you focus on the calls that need a human.

On every tool page in this directory, each item carries one of these four labels, so you know exactly what you are getting before you set it up.


Frequently asked questions

Is a Claude Skill the same as a ChatGPT Custom GPT? They are the same idea: a reusable instruction bundle you set up once and trigger again later. The naming and setup differ by platform, but both are skills, not plain prompts.

Is a long, detailed prompt a skill? No. A long prompt is still one-shot unless you save it as a named, reusable tool. The moment it is stored and triggers automatically, it becomes a skill.

What makes something an agent instead of a workflow? A workflow follows steps you designed in advance. An agent decides the steps itself to reach a goal, and keeps working after the initial instruction.

Do I need an agent to get value from AI in real estate? No. Most time savings come from skills and workflows on tasks you repeat. Agents help most on open-ended, multi-step jobs like full due diligence.

Mentioned in this post

Claude (Anthropic) logo

Claude (Anthropic)

General-purpose AI assistant from Anthropic with a large context window, used in real estate to read and Q&A long leases and offering memoranda, draft communications, and summarize market research.

Claude is a general-purpose AI assistant from Anthropic that reads long real estate documents and answers plain-language questions about them. Its flagship models carry a 1M-token context window, enough to hold a full lease, purchase agreement, or offering memorandum at once.

General AI Tools Research & Analysis llmassistantdocument-analysis
View profile
ChatGPT (OpenAI) logo

ChatGPT (OpenAI)

General-purpose AI assistant from OpenAI, used in real estate to write listing and marketing copy, draft outreach, summarize reports, and answer transaction questions.

ChatGPT is OpenAI's conversational large language model, and real estate teams use it to draft and summarize text and to spin up purpose-built assistants. Custom GPTs and built-in DALL-E image generation extend it to branded helpers and visual content.

General AI Tools Research & Analysis llmassistantlisting-copy
View profile
Perplexity logo

Perplexity

AI answer engine from Perplexity that searches the live web and returns cited answers, used in real estate for sourced market, comparable, and due-diligence research.

Answer engine that searches the live web and returns responses with numbered citations, so operators can fact-check every claim before acting. Useful for fast due diligence and CRE market scans where source-checking matters.

General AI Tools Research & Analysis answer-engineresearchcitations
View profile

Read next