Six Months In: Building a Thinking Partner · Post 4 of 4

What a Thinking Partner Is Made Of

Pete Griffin, Digital VisionOctober 2026

I set out to make Claude the perfect thinking partner. I spent a spring sprint testing nearly everything it can do, then three quieter months while the first versions were released and the apps sat in review, and had Claude count what that work still used.

This is what the partner turned out to be made of, in the order I'd set it up if I were starting again tomorrow.

1.A shared understanding, written down

One file that Claude reads at the start of every session. How I work. What I never want. How to check that something is done.

Mine has rules like "never claim a feature works without testing it" and "if you can't verify a step, say so plainly". It costs nothing and it's the first thing I'd write.

Start with five lines. Add one each time you have to explain the same thing twice.

2.Someone who questions the plan

Before any real build, Claude interviews me about what I'm asking for. One question at a time.

It was the most-used skill in the count, nine runs in three months. This is the closest thing to the partner I pictured. It catches the misunderstanding while it's still a conversation.

3.A second pair of eyes

When the work is done, a separate agent reviews it. Two thirds of the agent runs in the count were this: code review, security review, verification.

Whoever did the work is optimistic about it. A reviewer starting cold is not. That gap is where the problems show up.

4.Proof, not reports

Claude used a browser 486 times in those sessions, much of it loading the real page to look at what it had just built. I ask for the screenshot, the test output, the actual result.

"Done" is a claim. I wrote about that in June and I'd keep it word for word.

5.A count after every sprint

This is the new one. When a burst of experimenting ends, ask Claude to read your saved sessions and count what you still use. Put the rest away. Read your notes while you're there.

Mine came three months after the sprint. I'd do the next one sooner.

Keep dreaming, then prune

I don't regret a day of the sprint. Being inspired enough to try everything is what got me here, and some of what I tested will come back when a project calls for it.

The partner I wanted didn't arrive as one clever system. It arrived as four plain habits and a folder of notes, found by trying a great deal and keeping what the ordinary weeks used.

The rule: experiment freely, then prune. What's left is your thinking partner.

The numbers in this series all come from one audit of my own Claude sessions. Read the full report →

That's the series
Back to all 4 posts →