Download one file. Create two folders. Connect to an AI tool you already have. Drop in a meeting recording. Get structured notes with action items, owners, and deadlines. Total time: 18 minutes. Total cost: zero.
You've heard the pitch: AI can automate your workflows. But have you actually done it?
This article is the doing part. In 18 minutes, you'll have a working AI agent processing real input on your machine. Not a demo — a workflow you can keep using tomorrow.
We'll use the Meeting Notes Agent — because everyone has meetings, everyone takes notes, and almost nobody does it well.
What You'll Need
- A computer (Mac, Windows, or Linux)
- An AI tool — any one of these:
- 🟢 Google Gemini (recommended — supports auto-trigger on file arrival)
- 🔵 ChatGPT Plus (manual upload per session)
- ⚪ Claude (lightweight — conversational setup)
- A meeting recording or transcript (even 5 minutes of audio works)
No subscriptions. No installs. No API keys.
Step-by-Step: Deploy an AI Agent in 18 Minutes
Minutes 1–6: Setup
Download the blueprint. It's a structured Markdown document — an agent manifest that tells the AI exactly what to do:
| Blueprint Section | What It Defines |
|---|---|
| Trigger | File arrival in /input/ folder — audio, video, or transcript |
| Agent Rules | Extract attendees, decisions, action items; draft follow-up email |
| Agent Workflow | 8 steps: transcribe → metadata → topics → decisions → actions → summary → email → save |
| Tool | Gemini (primary), GPT / Claude (compatible) |
Create your folders:
meeting-notes-agent/
├── input/ ← Drop recordings here
├── output/ ← Agent writes notes here
└── agent.tact.md ← Blueprint fileThree items. That's your entire AI workflow infrastructure.
Minutes 7–12: Configure Your AI Tool
🟢 Gemini (Recommended): Open AI Studio → upload the blueprint as system instructions → point input to your /input/ folder. With a TACT runner, it auto-triggers on file arrival — no prompting needed.
🔵 ChatGPT Plus: Create a Custom GPT at chatgpt.com → paste Agent Rules + Workflow into instructions → upload files manually each session.
⚪ Claude: New conversation at claude.ai → upload blueprint + meeting file together.
Which to choose? Gemini for full automation. ChatGPT or Claude for a quick test.
Minutes 13–18: Run Your First Input
Drop in your meeting file — a recording (Zoom/Teams/Meet), transcript (Otter.ai, Fireflies), or typed notes. Place it in /input/.
Automated setup triggers immediately. Manual setup: prompt "Process the meeting using the Meeting Notes Agent blueprint."
Wait ~90 seconds. The agent transcribes → extracts → structures → saves.
What you get:
Meeting Notes: Q2 Marketing Planning
Date: Mar 28, 2026 | 47 min | 4 attendeesDecisions: Daily LinkedIn posting ✅ | $15K Q2 ad budget (60/25/15) ✅ | Case studies: 1/month from May ✅
| # | Action Item | Owner | Deadline |
|---|---|---|---|
| 1 | Daily content calendar | Sarah L. | Apr 4 |
| 2 | LinkedIn Campaign Manager setup | James T. | Apr 7 |
| 3 | TechCorp case study | Maria C. | Apr 10 |
| 4 | Q2 planning review | David K. | Apr 15 |
47-minute meeting → structured notes in under 2 minutes. Forward this to your team. That's your business case for the next agent.
Troubleshooting
| Issue | Fix |
|---|---|
| Missing attendees | Add names to input file or edit blueprint rules |
| Wrong action items | Agent extracts what was explicitly said — refine meeting habits or adjust rules |
| Garbled output | Use a transcript instead of raw audio |
| Format too long/short | Edit the tone or output format in the blueprint — it's Markdown, just edit it |
What You Just Did
- Downloaded one file
- Created 2 folders
- Connected to an AI tool you already had
- Dropped in a recording
- Got structured, actionable notes — zero manual work
18 minutes the first time. 5 minutes every time after. The setup is done — you just drop in the file.
This is what Article 1 meant by "workflows, not tools." This is what Article 2 meant by "your SOP is the blueprint." Now you've proved it.
More Agent Blueprints You Can Deploy Today
| Blueprint | What It Automates | Setup |
|---|---|---|
| Expense Receipt Processor | Extracts vendor, amount, category, GST from receipts | 15 min |
| Resume Screening Agent | Scores candidates, ranks shortlist against requirements | 18 min |
| Customer Inquiry Drafter | Classifies inquiries, drafts replies, flags escalations | 15 min |
| Quotation Evaluator | Weighted scoring across vendor quotes | 20 min |
| LinkedIn Copywriter | Generates posts following brand voice guidelines | 12 min |
One Agent Is Proof. Three Is a System.
The first agent is surprising — it works. The second is fast — you know the setup. By the third, something shifts: you stop following blueprints and start writing them. (This is exactly how 10 teams scaled from one agent to a system.)
The pattern — Trigger, Agent, Connector, Tool — shows up everywhere. In onboarding. In monthly reporting. In vendor negotiations. In the RFP responses that take your team three days.
Every one is a blueprint waiting to happen.
→ Describe your bottleneck. See it deconstructed into an agent blueprint.
Frequently Asked Questions
How long does it take to deploy an AI agent?
Using a structured blueprint, the first AI agent takes 18 minutes to deploy. This includes downloading the blueprint, setting up folders, configuring the AI tool, and processing your first input. Subsequent agents take 5-10 minutes.
Do I need coding skills to deploy an AI agent?
No. Agent blueprints are plain-language structured files. You need a computer, an AI tool (Gemini, ChatGPT, or Claude), and a meeting recording or transcript. No API keys, no installations, no programming required.
Which AI tool should I use for my first agent?
Gemini is recommended for full automation (supports auto-trigger on file arrival). ChatGPT Plus works well for manual uploads per session. Claude is lightweight for conversational setup. The blueprint works identically across all three.