Compound Leverage: THINK School Edition

This newsletter issue contains special THINK School content.

Enjoy this THINK School edition, brought by our sponsor, Morning Brew. Join over 4 million subscribers receiving daily business news for free.

If you want to support THINK School content, you can donate here.

Conversing with a potential client completely changed how I approach AI implementation with THINK.

"We love what your AI Scout does," they said during our call, "but adding another tool means changing our entire workflow, and we want our data to stay in an environment we control."

That got me to thinking

It wasn't about the technology—it was about friction.

They already had systems that worked, and my SaaS solution, no matter how powerful, meant disrupting established processes.

So I asked myself: what if I could give them and others like them the same functionality without a new online tool to learn?

I used Claude to build an agent that runs on anyone’s machine.

No new platforms to learn.

No data migration headaches.

No subscription fees.

Just an AI solution that works exactly how you already work.

I explained to Claude exactly what I wanted:

"I want to build a local version that allows people to upload contracting opportunities, analyze them, make recommendations based on that analysis, store results in organized files, and send out the findings."

From there, I created a folder on my desktop and had Claude build everything locally.

Breaking Down Your Manual Process with THINK

Here's where the THINK method becomes crucial:

T - Task: Map out every step you currently do manually
H - Hypothesis: Identify which steps an agent could automate
I - Invest: Build the agent piece by piece
N - Network: Connect the automated steps into a workflow
K - Knowledge: Test, refine, and scale the solution

For contract research, your manual process probably looks like:

1. Search for opportunities on sites

2. Download relevant opportunities or entering them in a spreadsheet

3. Review each contract manually

4. Score opportunities based on your criteria

5. Enter promising leads into your CRM

6. Follow up on deadlines

Each of these steps can be automated.

The Process

Step 1: Map Your Current Workflow I told Claude: "Walk me through building an agent that processes contract data. I download files from contracting websites, analyze them for fit, score opportunities, and create reports."

Step 2: Define Your Scoring System "Here's my scoring methodology: contracts worth $ 50 K-$500 K get priority, geographic preference for the East Coast, and must match these industry keywords."

Step 3: Specify Output Requirements "I want the analysis stored in CSV files in a desktop folder. Include cost estimates before running. Add filtering criteria and show me the result counts after processing."

Step 4: Choose Your Tools. Use Claude's API for the scoring analysis. Create a local web interface I can run from my desktop. Make it work with my Windows machine."

The Magic of Local File System Access

This is where it gets powerful. The agent can:

• Access folders on your desktop

• Read and write files directly

• Connect to other applications through APIs

• Send emails and notifications

• Run on schedules you set

It's like having a personal assistant that never leaves your computer.

Here's how I walked Claude through creating the agent:

*"I want you to build me a prompt that creates an agent to run on my machine. Here's what it needs to do:

1. Download contract files from X website

2. Analyze using my scoring criteria

3. Filter results based on my requirements

4. Output to CSV files in a desktop folder

5. Estimate costs before running

6. Show me filtered result counts Can you create the folder structure and all necessary files?"*

Claude responded with a complete setup including:

• Python scripts for each function

• Folder structure for organizing files

• Configuration files for my criteria

• Simple web interface to run everything

My first version had no fancy interface. I ran everything from the command line.

Download and save the file → Run analysis → Get results.

Once that worked, I added features:

• Automatic file downloads

• Scheduled processing

• Email notifications

• Better reporting

Traditional AI solutions force you to adapt to them. Local agents adapt to you.

You keep working exactly how you work now.

The agent eliminates the tedious parts.

Ready to try this? Here's my challenge:

1. Identify one manual process you do regularly

2. Download Claude for desktop or get an API key

3. Map out each step of your current process

4. Start with this prompt: "Help me build a local agent that automates [your process]. I want it to work with my existing workflow and output results in [your preferred format]."

Start simple. Get something working. Then improve it.

For contract analysis: "Build an agent that downloads contract files, scores them against my criteria [list criteria], and outputs ranked opportunities to a CSV file."

For lead research: "Create a local agent that takes company names from a spreadsheet, researches each one, and compiles findings into organized reports."

For content analysis: "Build an agent that processes uploaded documents, extracts key insights, and generates summary reports with action items."

Every successful local agent follows this pattern:

1. Input: What data goes in (files, URLs, spreadsheets)

2. Processing: What analysis happens (scoring, research, comparison)

3. Output: What results come out (reports, files, notifications)

4. Integration: How it fits your current workflow

Once you're comfortable with basic agents, you can build more sophisticated versions:

• Scheduled automation that runs while you sleep

• Multi-step workflows that chain processes together

• Integration with your existing tools and databases

• Custom interfaces tailored to your exact needs

You don't need coding skills to build powerful AI agents.

You must understand your process and know how to communicate what you want.

Local agents aren't just about avoiding subscription fees. They're about creating solutions that fit how you work.

Ready to eliminate some manual work from your day?

Start with a simple process.

Map it out.

Build an agent.

Test it.

You might be surprised how much you can automate with just a few well-crafted prompts.

This is exactly how you go from the THINK process to building something significant.

Start using THINK to assist you, then reach the crucial Invest stage, where you create the AI solution.

Real estate agent? Build an agent that finds property listings, analyzes market data, and identifies investment opportunities.

Consultant? Create one that researches prospects, analyzes their challenges, and generates tailored proposal outlines.

Accountant? Build an agent that monitors tax law changes, identifies client opportunities, and flags compliance deadlines.

The beauty is in what happens next.

Once your agent processes the data and identifies opportunities, it can trigger other actions: adding prospects to email campaigns, scheduling follow-up reminders, updating your CRM, or alerting you to time-sensitive opportunities.

Suddenly, you're not just eliminating manual work but creating velocity.

Your business will start moving faster because the research, analysis, and initial outreach will happen automatically while you focus on closing deals and serving clients.

I've built a government contracting agent called "Sam.Gov Agent" using this exact process.

It automatically finds federal contract opportunities, scores them against custom criteria, and generates actionable reports running locally on your machine.

Want to try it out for free?

All you need to get started:

• Download the files (I'll send them)

• Quick setup process (takes 10 minutes)

• Get a Claude API key (free to start)

• That's it

If you reply with "I want the SAM.Gov Agent," I'll send you everything starting next week.

Don't do government contracting? 

No problem. Sam still shows you exactly how local agents work, and if there's enough interest, I can build a version for non-government contractors.

Think about it - what manual research process could use its own "Sam" agent?

The best part?

Once you see how this works, you'll start spotting automation opportunities everywhere.

Ready to reclaim some hours in your week? Try creating your agent.

If you want to see a video demonstration of how to do this, let me know.

Marvin

News you’re not getting—until now

Join 4M+ professionals who start their day with Morning Brew—the free newsletter that makes business news quick, clear, and actually enjoyable.

Each morning, it breaks down the biggest stories in business, tech, and finance with a touch of wit to keep things smart and interesting.

Check it out