Running 18 Flips with Agentic AI: Real Use Cases From the Trenches

TLDR
Agentic AI can use a browser, read files, and carry a workflow across several steps instead of only answering a prompt. Ross uses it around 18 to 20 active projects, but he supplies the rules, reviews the output, and keeps sensitive actions behind approval.

Table of Contents

What Agentic AI Changes

The older use of AI was a chat. Ross could give ChatGPT or Claude his after repair value rules, provide an address, and ask it to search for comps.

Agentic AI can do more than return an answer. It can control a browser, read a folder, move through several steps, and run on a schedule. Ross names Claude Cowork, Claude Code, Perplexity Computer, and OpenClaw as current examples in the recording.

The tool is not the business brain. Ross still decides what the workflow should do and checks the result.

Deal Sourcing and Comping

Ross built a deal bot around Property Radar. Each month, it opens the site, builds the lists he defined, removes duplicates, exports the records, and sends them to his mail house.

That automation only works because the rules came first. Ross had already defined the list filters, buy box, duplicate logic, and handoff. The bot follows that system. It does not invent a good list on its own.

He still uses AI to help comp houses. The agentic version can search more places, but the value rules remain Ross’s rules.

Project Tasks, Bids, and Vendors

Ross manages about 18 to 20 projects. Instead of typing every update, he holds a spoken meeting with the AI. He talks through each property, who owns each task, and what is blocked. The system puts the work in the right project and builds a report.

Recorded calls can enter the same flow. The AI transcribes a call, finds the tasks, assigns them to the right project, and prepares the next message. Ross approves before it sends.

The bid and ledger workflow works like this:

  1. Ross creates a rough scope of work before he buys.
  2. He sharpens the scope after closing.
  3. Contractors send bids by text, image, email, or document.
  4. AI files each bid against the right project and contractor.
  5. Approved payments reduce the open bid balance in the ledger.

For vendor intake, Ross can photograph the name and number on a contractor’s truck. The AI adds that lead to his vendor database and trade category. This supports the depth chart instead of letting the contact vanish into his camera roll.

Mail, Property Management, and Contracts

Ross scans incoming mail to a Google Drive folder. The AI reads the PDF and routes it to the right property folder. He uses the same idea for job-site photos and videos.

He also uses AI to help manage the property manager. The review includes:

  • maintenance totals across the portfolio;
  • lease amounts compared with rent collected; and
  • upcoming vacancies.

Ross used to do that manual pass weekly and sometimes several times a week. Now the AI prepares a list so he can ask focused questions. It does not remove the owner’s duty to verify.

For acquisitions, the AI can match a recorded seller call to the CRM, find the property and contact, fill Ross’s contract template, and send it to the local printer. Ross still reviews the deal and the document.

Scopes From Video

This is the workflow Ross says he uses most.

He records a walkthrough and says the work out loud: appliances, trim, counters, rooms, and repairs. He then moves the video to his computer. The AI reads the words, uses his scope template and stored pricing, and drafts an itemized scope and budget.

It can only include what the system can see or what Ross says. The output still needs review. The transcript gives no measured before-and-after time.

Video Production Without an AI Script

Ross was also building a video workflow. After recording, the agent would:

  • remove silence;
  • normalize the audio;
  • read the transcript;
  • create chapters, a summary, and timestamps; and
  • upload and schedule the video.

There is one hard line: the AI does not write Ross’s script. He promises his audience real talk from his own career, not phrases generated by a model. AI can handle the production pipeline without replacing his voice.

The Foundation and the Security Limit

Ross says agentic AI works only after the foundation exists:

  • clear folders;
  • clear rules;
  • repeatable inputs;
  • known project and vendor records; and
  • human approval for sensitive actions.

He built a bot to answer Facebook Marketplace rental leads, then shut it down. A public message could contain a prompt injection that tries to make the bot reveal or misuse access. That risk was not worth it.

The durable rule is simple: organize the business first. Then automate a workflow you already understand. Keep public-facing and sensitive actions inside clear guardrails.

FAQ

Which tools did Ross name?

Claude Cowork, Claude Code, Perplexity Computer, and OpenClaw. The exact products may change, but those are the four examples named in the source.

Do I need to code?

Ross says much of the setup feels more like managing a person than writing code. You still need to explain the rules, folders, inputs, and approvals clearly.

How long does setup take?

The source gives no build timeline. It says the pre-work is real because each useful workflow needs a sound system underneath it.

Where should I start?

Pick one repeated task you already understand. Write its rules, clean the files it uses, define what needs approval, and test the result before giving it more access.