Tool
Complexity mode
Claude, without the five‑step plan.
Claude is very good at questions with a right answer. Maths, engineering, science, code. A lot of its training rewards answers a machine can check.
People bring it other questions too. Why a team won’t change. What to do about a boss, a friend, a family. Nobody can say in advance how those will go, because people respond to what you do. There is no right answer to find.
Claude doesn’t notice the difference. It answers with the same confidence, a tidy plan with numbered steps. Complexity mode makes it say it can’t know, then help anyway, with a few small things to try, cheap enough to drop if they go wrong.
“My co-founder keeps missing deadlines and I’m getting resentful.”
Get clear on the pattern before you talk. Have a direct, private conversation. Fix the system, not just the person. Set consequences and follow up. Check your own part.
Claude, no instructions
“I can’t know how this will go, and neither can anyone else. Two people adapt to each other all the time, and your resentment is already changing how you act around them, even if you haven’t said a word.” Then four small moves, and what to watch for.
With complexity mode
Try it
Copy the instructions and paste them as the first message of a new chat, in Claude or ChatGPT. Then ask what you’re stuck on.
For deeper answers
- Chapters
The book chapters it leans on. About 15,000 words with the notes below, so they use more of your limit.
- Notes
My notes on Samir Rihani’s work. Attach both files to the same chat.
- Claude Code
Download the skill, unzip it into
~/.claude/skillsand start with “use complexity-mode”.
Where it comes from
My book, The Pigeon Test: Why Change Fails, and Samir Rihani’s Complex Systems Theory and Development Practice. He taught me this at Liverpool. The notes are my adaptation of his work. The ideas are his.
What it can’t do
It can’t make Claude understand complexity. It stops Claude pretending to know how people will respond. I’ve only tried it on single written questions, and nobody outside my own testing has used it on a real problem. If you do, tell me how it went on LinkedIn. The call stays yours.
Sources for the opening: training on checkable answers, models sounding more certain than they are, people bringing Claude personal problems. It’s free, and it runs on your own account.