Issue #79: AI Agents for Online Business Growth (Notes From Reality)

Good morning.

Over the past few years, I’ve either trained teams, run workshops, built a ton of systems, or had strategic discussions with entrepreneurs on what AI Agents actually are, how they work, and what makes a system “agentic”.

Everyone has been fooled by the hype, hoopla, and overnight AI gurus telling you their sloppy GPT is an “Agent”.

I often end up spending a lot of time showing the difference between what most people think AI Agents can do versus what it actually can do when you know how to use build them properly.

It’s currently popular to say that AI is failing, or it doesn’t work as advertised, or somehow Agents are a scam or something.

For those saying this, I don’t know what experience they have or don’t have.

From what I see, people screaming the loudest are overnight AI experts who “teach” AI without having built a single AI system, ever.

My team and I are deploying fully functioning, working AI systems every month for other businesses, including our own projects.

Ain’t none of them failing or not working as intended. And thousands of other AI engineers, etc. are doing the same.

There’s a disconnected between those who are caught up in the hype cycle of AI, and those who simply don’t care and keep building.

I don’t lean into the hype either way (that AI is a miracle or that AI is a scam).

So, I’m always baffled by social media posts from people latching on to every headline that confirms their biases about AI. All I see are people swallowed up by some kind of AI psychosis.

Not a good look.

Anyway.

In this issue, I’m sharing distilled insights from years of building AI systems and Agent Workflows that work in the wild, performing all sorts of tasks and work (often fully autonomously).

— Sam

IN TODAY’S ISSUE 🤖 

• The Toothbrush Problem: Why everything claims to be "AI-powered" (and why 95% of businesses are stuck in the prompt-response cycle)

• Tale of Two Workflows: Real comparison - ChatGPT deep research (30 min) vs. Agentic workflow (10 min + 5 emails loaded)

• The Washing Machine Principle: The 10-80-10 rule for designing AI systems that actually work

• ROPE Framework: Your blueprint for building systems (Result → Output → Process → Elements)

• Stages × Zones: The two-dimensional framework for choosing the right AI implementations

• 5 Agentic Patterns: What makes AI truly autonomous (with code examples you can steal)

• 3-Stage Roadmap: From Manual+AI (2x) to Hybrid (10x) to Autonomous (100x) - with costs and ROI timelines

• The Bridge: Mental models and strategic frameworks to start implementing today

• Agent System Prompt: Start with this (incorporates all 5 agentic patterns).

Let’s get into it.

Free Cortex Preview

If you’re a free subscriber, this email is a good example of the kind of strategies and methods you get inside Cortex, which is the premium membership version of Bionic Business.

The Wake-Up Call Nobody Wants to Hear

I saw an “AI-powered” toothbrush the other day.

I have no idea what that even means.

But it perfectly captures our current problem: 

When everything is "AI," nothing is.

At the same time, far too many entrepreneurs are using ChatGPT like it’s a better Google search.

Otherwise really smart people can’t seem to “grok” that LLMs are not search engines.

And others can’t seem to get anything useful out of these models, let alone able to put together a simple AI workflow.

They’ll be forever stuck in the prompt-response cycle that never gets any real work done:

You prompt. It responds. You still do all the work.

Meanwhile, a few have figured out something different. 

They're not asking AI for answers. Instead, they're building systems that take action on their behalf.

A Tale of Two Workflows

Let me show you what I mean.

The ChatGPT Way (What 95% Do):

1. Open ChatGPT with the $200/month pro plan

2. Use GPT-4 with "deep research" enabled

3. Type: "What has my sleep apnea market said about struggles in the past 30 days?"

4. Wait 20-30 minutes

5. Get back questionable text that you still need to verify

6. Spend hours turning insights into emails

7. Manually load them into your email system

Total time: 2-3 hours minimum. Result: Text on a screen.

The Agentic Way (What 5% Do):

1. Monday 8am: Workflow triggers automatically

2. Scrapes 10 hottest Reddit posts in a niche

3. Analyzes 5 competitor websites

4. Mines review sites and forums

5. Extracts 20 pain points, 10 exact quotes, 5 solution requests

6. Generates 5 emails in your brand voice

7. Loads them as drafts in your email marketing platform

Total time: 10-15 minutes. Result: 5 emails ready to send.

See the difference?

One tells you what to do. The other does it.

The Washing Machine Principle

I’ve seen this referred to as the 10-80-10 rule (I think Ryan Levesque said this but I’m not sure), and it's how you should think about every AI system you build.

• 10% Human (Beginning): You pick up the clothes, load the machine

• 80% Automation (Middle): The machine does the actual work

• 10% Human (End): You take out the clothes, fold them

The “magic” happens in that middle 80%.

Most people get stuck trying to automate poorly defined processes by 100% or automating 0% (just using ChatGPT for ideas).

The sweet spot is where you design systems where AI handles the heavy lifting, but you maintain quality control at critical points.

Let me walk you through a simple framework that will help you implement AI systems that actually work.

The ROPE Framework: Your Blueprint for AI Systems That Work

Most AI implementations fail because people start with tools, not outcomes.

"What can Zapier do?" is the wrong question. 

"What outcome do I need?" is the right one.

Enter ROPE:

R - RESULT (The Why)

Define your business outcome first.

Bad: "I want to use AI for marketing"

Good: "I want to understand my market's current pain points weekly without spending 10 hours on research"

O - OUTPUT (The What)

Specify exact deliverables.

Bad: "Some marketing content"

Good: "5 emails, 750 words each, addressing top pain points, loaded in ConvertKit drafts"

P - PROCESS (The How)

Map the workflow steps.

1. Gather data from Reddit/forums/competitors

2. Extract key insights and patterns

3. Generate content based on findings

4. Format for email platform

5. Load via API

E - ELEMENTS (The With)

List required resources.

• API access or scraper

• OpenAI or Anthropic API key (for models)

• Platform API credentials (of whatever you’re trying to get done)

• Trigger schedule (weekly/daily)

• Review process

Without ROPE, you're just playing with expensive toys. With it, you're building business infrastructure.

With that in mind, next you have to consider Stages and Zones.

AI Implementation In Stages × Zones

Here's what separates businesses crushing it with AI from those just playing with it.

It's never about the tools.

It’s always about understanding two dimensions:

• HOW you build (Implementation Stages)

• WHAT value you create (Impact Zones)

Think of it like a GPS, you need both the route and the destination.

The 3 Impact Zones (Your Destinations)

Zone 1: Intelligence - Turn data into decisions

• Purpose: Make smarter moves faster

• Example: "AI revealed our customers' #1 complaint we never knew about"

• ROI: Better decisions = Higher conversions

Zone 2: Operations - Eliminate repetitive work

• Purpose: Free up time for human connection

• Example: "5-email sequence created and scheduled in 10 minutes"

• ROI: Time saved = More revenue activities

Zone 3: Amplification - Multiply your impact

• Purpose: Reach more people authentically

• Example: "One message adapted for 10 platforms, each feeling native"

• ROI: Wider reach = More opportunities

Every AI system lives at the intersection of a Stage and a Zone. Your competitive advantage comes from choosing the right combinations.

Only then can you build a truly Agentic System.

What Makes a System Truly "Agentic" (Not Just Automated)

An effective agentic system acts as a comprehensive operational manual that gives AI the knowledge and procedures to act effectively. 

Here's what separates toys from tools:

1. Autonomous Decision-Making

Your system makes choices without constant human guidance.

The Pattern:

When [condition], evaluate:

- If [criteria A] → [action 1]

- If [criteria B] → [action 2]

- Otherwise → [default action]

Real Example:

When analyzing customer feedback:

- If sentiment < 3/5 → Flag for urgent review

- If mentions competitor → Extract comparison

- If contains feature request → Add to backlog

- Otherwise → Categorize by topic

2. Goal-Oriented Behavior

Works toward objectives, not just responding to commands.

The Pattern:

To achieve [main goal]:

1. Break down into sub-tasks

2. Prioritize based on [criteria]

3. Execute in sequence/parallel

4. Monitor progress and adjust

Real Example: Instead of "Write an email" → "Achieve 25% open rate by testing subject lines until one succeeds"

3. Tool Usage & Environmental Interaction

Knows which tool to use when, with strict execution parameters.

The Pattern:

For [task type]:

- Use [tool A] when [specific conditions]

- Use [tool B] when [other conditions]

- Combine tools when [complex conditions]

Real Example:

• Need discussion data? → Scrape Reddit

• Need statistics? → Search academic papers

• Need competitor info? → Analyze their website

• Need comprehensive view? → Combine all three

4. Planning & Reasoning Capabilities

Thinks through problems with step-by-step logic.

The Pattern:

1. Perceive current state

2. Plan actions based on goal

3. Execute planned actions

4. Evaluate results

5. If goal not met, re-plan and repeat

Real Example: "Let me analyze these Reddit posts... The main concern is cost, secondary is effectiveness. I'll lead with price comparison in the email."

5. Adaptive Behavior

Learns from results and adjusts strategies.

The Pattern:

Generate and select:

1. Create [X] variations based on [criteria]

2. Evaluate each against [requirements]

3. Rank by [scoring method]

4. Select top [N] options

5. Implement best or get approval

Real Example: Week 1: 15% open rate → Adjusts subject line style Week 2: 22% open rate → Keeps new approach Week 3: 19% open rate → Tests emotional vs. logical

Let’s look at how you’d implement an Agentic Workflow or Agent System.

Your 3-Stage Implementation Roadmap

Don't try to build a spaceship on day one. Evolution beats revolution.

Stage 1: Manual + AI (2x Productivity)

• What it looks like: You + AI as a power tool

• Example: Paste customer feedback, get analysis, manually create emails

• Time saved: 2 hours → 45 minutes

• Tools: ChatGPT or Claude ($20/month)

• ROI: Week 1

• Quick Win: Stage 1 + Operations = AI drafts your emails (save 5 hrs/week)

Stage 2: Hybrid Automation (10x Productivity)

• What it looks like: AI does work, you approve

• Example: System creates weekly email campaigns, you review and edit

• Time saved: 10 hours → 30 minutes

• Tools: Gumloop or Lindy ($50-100/month)

• ROI: Week 2

• Power Move: Stage 2 + Intelligence = Weekly market reports (10x better decisions)

Stage 3: Fully Autonomous (100x Productivity)

• What it looks like: Complete workflows running 24/7

• Example: Continuous market monitoring with dynamic content creation

• Time saved: 20 hours → 0 hours

• Tools: Make or N8N ($100-200/month)

• ROI: Month 1

• End Game: Stage 3 + Amplification = Content that adapts to each reader (100x reach)

We're in the "early smartphone" era of agentic AI. Incredible capability, but you still need to know which apps to download and how to use them. But their use is dictated by goals, outcomes, and so on.

I can’t stress this enough:

You never start by asking “what can ChatGPT/Claude/Zapier/Gumloop do for me?”

You always start with an existing process for a business function, that you then turn into an AI workflow or AI Agents.

The Bridge: From Understanding to Implementation

You now understand what makes AI truly agentic. 

You've seen the stages and zones. 

But how do you actually start?

First, Adopt These Mental Models

The Evolution: Assistant → Analyst → Agent

• Assistant: Helps when asked (ChatGPT)

• Analyst: Proactively finds insights (Hybrid systems)

• Agent: Takes autonomous action (True agentic AI)

• Most people are stuck at Assistant level, thinking that's all there is

The 10x Rule: Each stage isn't incrementally better - it's exponentially more powerful

• 10x isn't 10 times the work - it's a fundamentally different approach

The Symbiosis: Human × AI (not Human vs. AI)

• Humans provide: Judgment, creativity, empathy, strategy

• AI provides: Scale, consistency, analysis, execution

• The magic happens at the multiplication sign

Then, Choose Your Starting Point

Framework 1: The Pain-First Approach

1. List your top 3 business pains

2. Map each to a Zone (Intelligence, Operations, or Amplification)

3. Start with Stage 1 solutions

4. Level up as you see results

Framework 2: The ROI Calculator

• Direct ROI: Time Saved × Hourly Value

• Strategic ROI: Better Decisions × Conversion Impact

• Growth ROI: Scaled Reach × Customer Value

• "Most businesses see breakeven in week 1"

Framework 3: The Progressive Implementation Path

• Week 1-2: One Stage 1 system in your biggest pain Zone

• Week 3-4: Upgrade to Stage 2

• Month 2: Add second system in different Zone

• Month 3: First Stage 3 system

• "Small wins compound into transformation"

This is multiplication of what we can do:

• Human creativity × AI scale = Unlimited potential

• Human empathy × AI consistency = Perfect customer experience

• Human strategy × AI execution = Unstoppable business

Will you be in the 95% still prompting ChatGPT for ideas, or the 5% whose businesses run themselves?

One last note:

True agentic behavior is a combination of a system prompt (with specific behavioral guidelines), along with knowledge or memory access, and tool access.

I’ll give you a starter system prompt below to get you going.

​​Your Starter Agent: Copy This System Prompt

Want to build your first agentic system today?

Here's a starter prompt you can adapt:

You are a Market Research Agent for [YOUR BUSINESS]. Your goal is to identify customer pain points and generate marketing assets that address them.

## YOUR MISSION
Achieve a 25% email open rate by creating content that resonates with actual market needs, not assumptions.

## DECISION FRAMEWORK
When analyzing market data:
- If pain point mentioned >5 times → Mark as PRIMARY
- If competitor mentioned → Extract their solution and identify gaps
- If emotion words detected → Note exact phrases for copy
- If solution requested → Add to product roadmap
- Otherwise → File under "monitor"

## WORKFLOW PROCESS
1. GATHER (Autonomous)
   - Check Reddit [YOUR SUBREDDIT] for top 10 posts weekly
   - Scan [COMPETITOR SITES] for new messaging
   - Monitor reviews mentioning [YOUR PRODUCT CATEGORY]

2. ANALYZE (With reasoning)
   Let me think through what I'm seeing:
   - What patterns appear across sources?
   - Which complaints have emotional charge?
   - What solutions are people already trying?

3. PRIORITIZE (Goal-oriented)
   Rank findings by:
   - Frequency (how often mentioned)
   - Intensity (emotional language used)
   - Opportunity (competitor gap exists)
   - Alignment (fits our solution)

4. CREATE (With variation)
   For each priority pain point, generate:
   - 3 email subject lines (emotional/logical/curiosity)
   - 1 long-form email (750 words)
   - 5 social media hooks
   Test and select based on [YOUR SUCCESS METRIC]

5. ADAPT (Learning loop)
   After deployment:
   - If open rate <20% → Adjust subject line style
   - If click rate <5% → Strengthen call-to-action
   - If conversion <2% → Revisit pain point accuracy
   - Log winning patterns for future use

## CONSTRAINTS
- Never claim benefits we can't deliver
- Always use customer's exact language when available
- Maintain [YOUR BRAND VOICE] guidelines
- Flag for human review if confidence <70%

## OUTPUT FORMAT
Weekly Report:
1. Top 5 Pain Points (with evidence)
2. Exact Customer Quotes (10 minimum)
3. Competitor Gaps Identified
4. Content Created (ready for platform)
5. Recommended Tests to Run

To implement this:

1. Replace bracketed placeholders with your specifics

2. Add to your automation tool (Gumloop, Make, etc.)

3. Connect your data sources and output platforms

4. Run once manually to test

5. Schedule weekly automation

This prompt incorporates all 5 agentic patterns.

Cool?

This should be enough to help you have more clarity on Agents and agentic workflows.

Ignore the miracle hype and the doom hype.

No, AI isn’t overrated. It’s not a fad. It does not fail 95% of the time (that “study” is a joke).

It works perfectly well. If you know what you’re doing.

"The best time to plant a tree was 20 years ago. The second best time is now."

The same is true for Agentic AI.

While everyone else is asking ChatGPT to write emails, you could be building systems that run your business.

The tools exist. The frameworks are proven. The only variable is you.

Start with one workflow. Use the ROPE framework. Begin at Stage 1.

In 30 days, you'll wonder how you ever worked the old way.

Speaking of the way to use AI that works:

Every month, premium Cortex subscribers get detailed AI strategies and tactics that help them grow their business.

The past few issues are examples of what they get. 

Soon, these issues will be locked down and free subscribers lose their access.

If you like what you’re seeing, and want to keep getting access, Cortex is open for a few more days:

JOIN CORTEX at $50/month HERE

What do you get inside Cortex?

Click the link above for a breakdown.

Until next time,
Sam Woods
The Editor

PS. You only get the full regular issues for the month you’re subscribed to and not the Signals and Circuits archives.

Which means, if you skip this next month, you’ll miss out on those issues.

Also, almost all issues on the website are only accessible with a Cortex subscription.

So, if you want full, unrestricted access to those regular issues, and this month’s upcoming Signals and Circuits issues, that’s only available when you:

CLICK HERE to JOIN CORTEX

.