Starting today, I'm going to share how I use the THINK system: what's working, what's failing, and what I changed.
Take what fits. Learn from my mistakes. Apply it to a goal you’re working towards.
My goal is a growth target over the next fifteen months. Getting there required changing how I think and what I do every day. I'm not there yet. That's exactly why I am sharing everything so you can learn from my mistakes.
How the system works, in one minute
THINK runs on four intelligence pillars, each staffed with agents and agentic workflows:
Score finds, evaluates, and prioritizes. It watches for events, scores the opportunities, and monitors for changes each week.
Intelligence turns signals into output. This is where sales and marketing activities happen. For example, I have agents that do content research and hand me a report on what's trending and why.
Execution extracts the value. This is where sales, fulfillment, and operations happen. Workflows I need to create value from Score and Intelligence work.
Signal keeps the watch on cadence and reports back.
Every agent's instructions are written across five libraries I use. The rules they follow all live in one place, so when something changes in the market or my approach, I update the rule once, and every agent follows it from then on. When I fix a mistake, I can update it without rebuilding anything.
I didn't start with this structure. I started running it differently and landed here through trial and error as new data came in.
It will look different again by December. That's the point. The thinking happens in Claude, Gemini, and Perplexity.
The agents run on schedules in Claude Code; a Codex agent is responsible for checking and testing their work and handling documentation.
One agent sits above the rest with her own email account. She oversees the others, then reports to me, flagging what needs attention and action on my part.
The design principle: work comes to me for judgment. I don't chase it.
What's working
Every Monday, while I sleep: 4 AM, one agent maps where capital is clustering. 5 AM, another runs the radar. 6 AM, a master report compiles. At 9 AM, Sarah, my business growth manager, emails me the result. Sarah is an agent.
On May 25, her report opened like this:
"Qualified pipeline is flat for 14 days and we lost 162 newsletter subscribers this week, but 7 high-value capital events landed this run, including a July 10 federal deadline that requires a move from you in the next 10 days."
The bad news, the good news, and the deadline that needs my judgment. Three lines. Inside that report: a federal program worth $1M to $8M per grant with a July 10 deadline, why we're positioned to lead the regional application, and a one-line diagnosis of this newsletter: "The content is working. The problem is cold acquisition, not engagement."
Detection is the front half. Every opportunity lands in an inbox. I review and qualify; another agent picks them up, and I flag each one as partner or solo.
The ones my partner and I review together to see if we can build something real. Survivors get entered into campaigns. The agents move opportunities to the table. Humans still decide what's worth building and pursuing.
Today, I hold $8,027,250 in documented pipeline.
Every dollar traces back to something the system surfaced first.
What's not working
The list lost 162 subscribers that week. The system caught it and told me where the problem isn't. It can't tell me why people leave. I'm auditing that now.
But the biggest failure in May wasn't the system. It was me. The agents researched, built the content strategy, and notified me to review and convert it into posts. That's where it stopped: zero of nine planned posts published, four site pages past deadline, everything sitting there finished, waiting on me. The system did its job. The human was the bottleneck. Scale ends at the part you do not take action on.
And while reviewing agent notifications, I noticed I hadn't received any new entries for two weeks, so I had to adjust how it was running to get results.
The system that watches everything still needs someone watching it. It’s where I got the idea for Codex agents to check the system. Agents that prompt themselves, instead of waiting on me, add the oversight layer that the system was missing.
What changed
Two decisions: my conversion step now gets a standing block on my calendar, the same way the agents have theirs, and two new deals get dollar amounts instead of me admiring the total.
The lesson this month: the system watches, scores, researches, and reports. My job is judgment and follow-through. When I don't show up for my part, the system's output is just unfinished homework I didn't do. I run the whole system with tools I already use and free ones.
How to build your own version?
I did not build this all at once, and you shouldn't either.
The first step is your Foundation: a playbook that finds the signal, your target outcome, your blindspots, and what you should be watching for. Without it, you can't decide what to score, what to monitor, or which digital employee to build first. Then you move to Score, and the system starts doing that work for you.
If Foundation is where you are, I built a free tool for exactly this: the Signal Finder. Tell it your goal, and in about 10 minutes, it returns your next move and the beginning of your playbook. It runs inside Claude.
I will publish THINK Reports monthly throughout the summer. What moved, what broke, what I changed.
If you want it to keep going after that, reply and let me know. I read every reply.
Marvin