Breeding AI agents that navigate portals, book flights, and act like team members

AGENTIC AI BREEDER AWARD - PROJECT DESCRIPTION

Building AI That Thinks, Navigates, and Acts

The Shift

There's a difference between automation and agents.

Automation: When X happens, do Y. Agents: Figure out what needs to happen, navigate to where you need to go, make decisions along the way, and get it done.

Automation follows rules. Agents solve problems.

As Head of Organizational Development at Fifth Avenue Properties—a homebuilder scaling from 117 to 919 home starts this year, with a BHAG of 5,000—I've moved beyond simple automations. I'm breeding AI agents that work like team members: they log in, they navigate, they think, they act.

This is what I've built.

Agent #1: The Utility Provider Extraction Agent

The Problem: As a homebuilder, we have utility accounts across dozens of providers. Each provider has a different portal, different login, different interface. Extracting balances manually means logging into portal after portal, copying numbers, building spreadsheets. Hours of tedious work every month.

The Agent I Built:

Agent Activated

→ Logs into utility provider portal (knows credentials)

→ Navigates through each house in the account

→ Extracts account number and amount owing

→ Moves to next property, repeats

→ Compiles everything into spreadsheet

→ Accounting reviews for payment

→ Agent enters credit card and pays each one

→ Expenses automatically recorded

What Makes This Agentic:

- Autonomous navigation — The agent doesn't follow a fixed script. It navigates different portal interfaces, handles variations, finds the data.

- Multi-step reasoning — Login → navigate to account → find each property → extract data → move to next → compile → pay

- End-to-end execution — From login to payment, the agent handles the entire workflow

The Impact: What used to take hours of portal-hopping is now autonomous. The agent logs in, extracts, compiles, and pays—I review the results.

Tools: n8n, browser automation, AI agent

Agent #2: The Flight Booking Assistant

The Problem: Our CEO needs to book flights. The traditional process: search multiple sites, compare options, make a decision, go back and book, add to calendar, log the expense. Time-consuming. Distracting.

The Agent I Built:

CEO posts in Teams "Flights" channel:

"I need to fly to Vancouver next Tuesday, back Thursday"

→ Webhook captures the request

→ AI agent parses: destination, dates, preferences

→ Agent searches flight options

→ Agent compares prices, times, airlines

→ Agent asks clarifying questions if needed:

"Do you prefer morning or afternoon departure?"

"Direct flights only, or open to connections?"

→ Agent presents top options in Teams chat

→ CEO selects preferred option

→ Agent books the flight

→ Confirmation sent back to Teams

What Makes This Agentic:

- Natural language interface — CEO just says what he needs in plain English

- Clarifying questions — Agent asks for more information when needed, just like a human assistant would

- Decision support — Doesn't just dump options; compares and recommends

- End-to-end execution — From request to booked flight, all in Teams

The Impact: The CEO never leaves Teams. He states what he needs, answers a question or two, picks an option, and the flight is booked. A 30-minute task becomes a 2-minute conversation.

Tools: n8n, Microsoft Teams webhook, AI agent, browser automation

Agent #3: The Profile Picture Agent

The Problem: Building an executive KPI dashboard in Airtable. Wanted profile pictures for each executive. Could manually search the website, download each photo, upload to Airtable. Tedious.

The Agent I Built:

For each executive in the KPI table:

→ Agent navigates to company website

→ Searches for executive's profile

→ Extracts profile picture

→ Uploads to Airtable record

→ Moves to next executive

What Makes This Agentic:

- Autonomous browsing — Agent navigates the website on its own

- Search and extraction — Finds the right person, pulls the right image

- Batch processing — Handles multiple executives without manual intervention

The Impact: Small task, but proves the concept: agents can browse, search, and extract—just like a human assistant would.

Tools: Airtable agent mode

Agent #4: Email Draft Assistant (In Refinement)

The Problem: Email takes hours. Reading, understanding context, drafting responses. Necessary but draining.

The Agent I Built:

New email received

→ Agent analyzes email content

→ Agent assesses sender context and intent

→ Agent drafts response matching my communication style

→ Draft saved for review

→ I approve, edit, or rewrite

→ Human sends

What Makes This Agentic:

- Context understanding — Doesn't just respond to keywords; understands what the person actually needs

- Style matching — Drafts in my voice, not generic AI voice

- Human-in-the-loop — Agent drafts, human decides

Current Status: Under refinement. Core functionality works—drafts are being generated. Tuning the AI's understanding of context and tone to get closer to send-ready on first pass.

Tools: n8n, email API, Claude API

Coming Soon: Building Permit Agent

The Vision: When a new project is created in Airtable, an AI agent automatically fills out the building permit application.

The Architecture:

New project created in Airtable

→ Agent pulls project data (lot, specs, community)

→ Agent navigates to municipal permit portal

→ Agent fills form fields intelligently

→ Agent flags for human review

→ Human approves → Agent submits

Why It Matters: At 919 homes this year—and 5,000 at our BHAG—permit applications are a volume game. An agent that pre-fills 90% of the work transforms a bottleneck into a checkpoint.

Experimenting: Voice AI for Sales

The Exploration: I've been toying with ElevenLabs to create a voice AI agent on our website—one that can engage visitors, answer questions, and funnel them toward sales conversations.

The Vision: A prospect lands on the website at 10 PM. Instead of filling out a form and waiting, they talk to an AI agent that understands our homes, our communities, and our process. The agent qualifies them, answers their questions, and routes them to the right salesperson.

Status: Early experimentation. But it represents where this is heading—agents that don't just work behind the scenes, but interact directly with customers.

This Is Just the Beginning

I want to be honest: I'm still developing and acquiring ideas. I don't have all the answers yet.

The agents I've built are working. The ones in development are progressing. But there's so much more potential—agentic workflows I haven't even conceived of yet.

The constraint isn't vision. It's time.

This is all built on top of a full-time job as Head of Organizational Development. I'm not in a lab somewhere with unlimited hours to experiment. I'm scaling a homebuilder from 117 to 919 homes while simultaneously building the AI infrastructure that will get us to 5,000.

Every agent I build has to earn its place by solving a real problem, saving real time, or enabling real scale. That's the filter. That's why the ones I've built actually work in production.

But make no mistake—I will continue to push the envelope. As new agentic capabilities emerge, as new problems surface, as new ideas click into place, I'll keep breeding agents.

This is just the beginning.

The Agentic Toolkit

n8n — Workflow orchestration with AI agent nodes. Handles webhooks, API calls, conditional logic, and browser automation.

ChatGPT Agent Mode — For tasks requiring reasoning, research, and multi-step thinking.

Airtable Automations + Agent Mode — For data-driven agents that live close to the data.

SweetProcess — Producing SOPs with automation triggers.

Browser Automation — Letting agents navigate websites, fill forms, extract data.

Microsoft Teams Integration — Natural language interface for agent interactions.

ElevenLabs — Voice AI for customer-facing agent experiments.

What Makes an Agent Different

Automation Agent Fixed rules Dynamic reasoning Single action Multi-step navigation Trigger → Response Conversation → Decisions → Actions Follows a script Adapts to variations Needs exact conditions Handles ambiguity The agents I build don't just execute. They think.

The Utility Agent doesn't need every portal to look the same—it navigates whatever interface it encounters. The Flight Agent doesn't need perfectly formatted requests—it asks clarifying questions. The Email Agent doesn't need templates—it understands context.

Key Learnings

1. Agents are team members, not scripts — Think of them as junior employees who can browse, search, and act. What would you delegate to a capable assistant?

2. Human-in-the-loop for high stakes — Agents research and recommend; humans authorize. The Flight Agent books, but only after the CEO picks the option.

3. Natural language is the interface — The best agent interactions don't feel like using software. They feel like talking to a colleague.

4. Start with navigation + extraction — Before building complex agents, prove they can log in, navigate, and pull data. That's the foundation.

5. Clarifying questions change everything — An agent that asks "morning or afternoon?" is 10x more useful than one that guesses.

6. Tedious + repetitive = agent opportunity — If it's boring and someone does it monthly, an agent should do it.

7. Build agents that earn their place — When you're doing this on top of a full-time job, every agent has to solve a real problem. That constraint makes the ones you build actually work.

Relevant URLs

www.fifthavehomes.com, fifthaveproperties.com

Closing

I'm not just building automations. I'm breeding agents.

The Utility Agent that logs into portals, extracts balances, and pays bills. The Flight Agent that lives in Teams, asks clarifying questions, and books travel. The Profile Agent that browses our website and pulls executive photos. The Email Agent that reads, understands, and drafts.

These aren't scripts following rules. They're AI systems that navigate, reason, and act—like junior team members who never sleep.

We're scaling from 117 to 919 homes this year. Our BHAG is 5,000. At that scale, I can't hire enough humans to handle every repetitive task. But I can breed agents.

That's what I'm building. That's how we scale.