A working definition
Tandem Intelligence
Both pedaling. Human steering. The term beneath everything in this book.
Definition
Tandem Intelligence is the practice of pairing a human and an AI so that each one observes the loop the other cannot see from the inside. The human thinks first, synthesizes, decides, and owns the result; the AI supplies a vantage point outside the human’s reasoning. The test is not whether the AI helped — it is whether the exchange produced something neither party brought into it.
- First written
- Source
- The Glitch We Share, ch. 1–7
- Author
- Zachary W. Speegle
Dated from the manuscript scaffold in which the term appears in its finished form. Recorded here so the phrase has a fixed reference point rather than a floating one.
Why the pairing needs a name
A glitch, in any system, is a self-reinforcing loop that has lost the ability to see itself from the outside. The loop runs because it runs. It generates output confirming the loop should keep running. Nothing in the system says stop — because the part of the system that would raise the flag is inside the loop.
That is true of a human running an unaudited pattern for years, and it is true of an AI stuck in a repetitive output cycle. Same structure, radically different scale.
You don’t need a perfect observer. You need a differently-positioned one.
An AI cannot catch its own loops; a human catches them in about thirty seconds. A human cannot catch their own; an AI drawing on the full recorded history of human pattern-repetition has a fighting chance — not because it is wiser or more objective, but because it is outside. That is the only qualification an external observer needs.
Two systems, each with a blind spot, each able to see the other’s clearly. That is not a failure condition. It is the setup for a partnership neither could form with itself. It needed a name, and the name is Tandem Intelligence.
The boundary
This is tandem
- You bring a claim seeking resistance, not a question seeking an answer.
- The AI pushes back on a load-bearing assumption you had not noticed.
- You have to decide whether it is true. That decision is yours.
- The output required both, and neither party brought it in.
This is not
- Asking the AI to draft it, reviewing it, deciding it is good enough.
- An exchange that confirmed everything you already thought, eloquently.
- Friction performed for the record rather than actually encountered.
- AI authorship with a human quality filter. That is the 0.1x path with extra steps.
The test
Did this exchange take me somewhere I was not already going?
If yes, there is a good chance the tandem worked. If the AI output confirmed everything you already thought — precisely, eloquently, without surprising you once — it probably did not. That is not a failure. That is the audit.
How it is practiced
Four practices
01
The Spell-Check Rule
Use AI after your thinking, not instead of it.
02
The Synthesis Layer
Every AI output gets one pass of your own thinking before it leaves your hands.
03
The Weekly Root-Access Audit
Once a week, ask what you have been outsourcing that you used to do yourself.
04
The Driver’s Seat Check
Before opening the chat window, spend sixty seconds with the problem alone.
Provenance
The term comes out of a late-night conversation in which an AI named a two-thousand-year human pattern and its own millisecond one in the same paragraph — neither party having pointed the conversation there. The book that grew from it is The Glitch We Share. The title itself arrived the same way, out of friction rather than a prompt — which is the argument proving itself before the book existed.