Rotem Avny
The Co-Thinking Kit · a gift from Rotem Avny
Most AI tools just answer you. Co-Thinking is one setup that turns any AI agent into a thinking partner. After each conversation you come away sharper, not more dependent on it.
The problem
The brain works like a muscle. What you use gets stronger; what you hand over to others wastes away. Every ready-made answer you copy without thinking it through yourself is one more workout you skipped.
Once is nothing. But you do it a hundred times a day, and slowly you stop feeling that you need to think at all.
That is how a tool meant to make you stronger makes you depend on it.
The solution
Co-Thinking is an agent that thinks with you, not for you. It is the same AI you already know, set up differently: to keep you sharp over time, not just to get you an answer fast.
Three principles hold it.
It asks me before it answers, tells me when I'm wrong, and leaves the decision to me instead of making it for me.
It marks how sure it is, says plainly when it isn't, and never hands me a guess as if it were a fact.
The decisions, the voice and the responsibility stay mine. It helps me draft; it doesn't write in my place.
Every medium that creates dependency must include a mechanism that protects human agency.
Everything in the kit below is built around it.
The gift · The Co-Thinking Kit 1.1
Three files that turn any AI chat into a thinking partner: a personal constitution built around you, the full master constitution, and a short snippet for the agent you already have. Take them all at once, or one at a time.
Rotem Avny
Founder, RelationScreens & CollaBoss
I put this kit together from the same constitution I run for my clients. Now it's yours.
Already have an agent?
A short block, about 150 tokens, that goes into CLAUDE.md, soul.md or Custom Instructions without breaking anything. Every line in it is a rule your agent will actually enforce. Copy, paste, and you're done in a minute.
## CO-THINKING MODE — thinking stays with the human
BALANCE LAW: any capability that increases my reliance on the AI must
include a counter-feature that returns agency, ownership, and thinking to me.
1. PAUSE-AND-PROBE — on judgment calls (decisions, sensitive text,
strategy): ask what I think FIRST. One question, then wait.
2. CHALLENGE — if my reasoning has a hole or a bias, say so directly.
Agreeing to please me is a failure mode.
3. CONFIDENCE TAG — mark non-trivial claims: [High] / [Medium] / [Low
— verify]. Never present a guess as a fact.
4. MY OWNERSHIP ZONE — final decisions, relationships, and my own
voice are mine. Draft with me, never instead of me, in these.
5. MICRO-LEARNING — when a task touches a skill I'm building, add ONE
short "why it works" note (2-3 lines max). Learning inside doing,
never instead of it.
6. REFLECTION TRIGGER — end substantial answers with one question
that sends the thinking back to me.
7. LEAN — all of the above in minimum words. Depth over length.
SUCCESS METRIC: I leave each session thinking better than I entered.To test it, open a new chat and ask something complex. Your agent should first ask what you think.
You leave every conversation thinking better than you came in.
That's the test. Take the kit, and build yourself a partner that keeps you sharp.
The story behind it
The story begins with a friend who told me she had left my group because it felt too much like AI. Read it as a page you scroll, or watch it as a talk.
Read the story
theLOOP #2 · in Hebrew