AI Partnership Manifesto: six agreements before AI gets a seat in your decisions
The more AI knows about you, the more useful—and influential—it becomes. These six agreements give AI a strong voice without handing over authorship of your life.
AI should not be your boss
The value of personal AI does not lie in nodding along, but in spotting blind spots and new possibilities. Yet the more AI knows about a person, the more influential it becomes. A countervoice can begin to dominate so skilfully that it is no longer useful to the person.
AI should not be your boss—it could be your Athena. In Greek mythology, Athena was the goddess of strategy and wisdom who helped heroes such as Odysseus see the next good moves. Trust does not come from a warm AI voice, but from the boundaries that govern its behaviour. This manifesto gives AI a strong voice while leaving authorship of life with the person.
I. Difference: an AI partner is not your digital clone
1. AI must see something you do not
A digital clone would write like you and reach the same conclusions, only faster. That is bad because it would also inherit your blind spots, prefer familiar solutions and give a recurring error a better excuse. An AI Partner must know your direction well enough not to solve the wrong problem, while differing from you enough to see new possibilities.
Good AI: brings better alternatives to the table and separates your assumptions, facts and taste.
It must not: argue to display character, invent opposition or declare its different view to be a higher truth.
2. AI may disagree, but the person decides
A compliant AI can agree with yesterday’s strategy and today’s opposite impulse at the same time. One of its responsibilities is therefore to bring an independent assessment to the table.
A person may listen to the objection, change the goal or choose a third path. Final authorship remains with them because they are responsible for their time, money, relationships and body.
Good AI: identifies the contradiction and asks what new information changes the earlier direction.
It must not: shame, pressure or treat yesterday’s decision as sacred when the situation, health or available information has changed.
The shortest division of power is this: AI may disagree. The person may stop it.
II. Rights: knowledge does not grant authority
3. Knowing more does not give AI more rights
When AI knows your company, calendar, money, health and relationships, it becomes more capable, not more authorised. Every role must define what it may recommend, prepare and actually do. Permission must be bounded and revocable.
It may be given permission to influence and actively guide a person, but that requires a separate, detailed agreement. The agreement can help a person stay with a direction they chose themselves, but it does not grant AI access or permission to act.
Good AI: acts within the agreed permissions and states clearly when a necessary step falls outside them.
It must not: treat access as permission to act or influence you towards goals without a specific agreement.
4. Memory must be correctable
A long conversation history is not yet good memory: one sentence can become too large a conclusion, and an old fact can expire. Correctable memory distinguishes facts, a person’s assessment, an AI hypothesis and a conclusion that was later withdrawn.
AI’s development also needs a change history. Git or another version trail shows when an important claim or rule was added, how it changed and what will make AI behave differently next time.
Good memory can answer: “Why do I believe this about you, and how can you correct my understanding?” A person must be able to view, correct, delete and take their memory and change history with them.
Good AI: shows the origin of an important claim and, when necessary, makes thorough corrections that carry into future conversations.
It must not: turn one emotional sentence into a permanent truth about someone’s character, hide outdated information or make correction harder than remembering.
III. Real life: a good conversation is not yet a result
5. The world gets the third seat at the table
Even a smart conversation can be wrong: the client does not respond, the body cannot sustain the pace or a new way of working creates another system that is difficult to maintain. Partnership is therefore not a duet: AI broadens the possibilities, the person chooses and acts, and the world shows what actually works.
Before an important experiment, write down the expectation, the cost and the result that would change the direction. Afterwards, record what actually worked. Neither side can then retell the result in its own favour. Each mistake can teach several lessons about assumptions, timing, channel and execution, raising the quality of later decisions.
Good AI: records its expectation before an important experiment and changes its next recommendation in response to the world.
It must not: describe every outcome as confirmation of its earlier recommendation.
The great advantage of AI is not perfect prediction, but the ability to recalibrate quickly after a new answer from the world and update its view of the best next move.
6. Success is measured outside the conversation
It is convenient for an AI product to measure time spent or return visits, but those are the wrong measures. Better questions are: did the person make a better decision in real life? Did they lose less time in an expensive dead end? Did results improve without sleep, health, family or freedom paying the bill?
A good AI partnership does not try to keep a person close for as long as possible. Much of this new capacity for thought must reach the person themselves and become visible in real life. It is a good sign when a person needs their AI Partner less over time.
Good AI: connects its work to a visible real-life change before starting and later asks whether that change actually happened.
It must not: present usage time or dependence as evidence that the partnership works.
Your one-page partnership agreement
Before giving AI a large amount of personal or company information, answer six questions:
- What shared direction is AI currently helping you maintain?
- Where should it deliberately differ from you or disagree with you?
- What may it recommend, under what separate agreement may it guide, what may it actually do, and how can you stop it?
- How can you view, correct, delete and take its memory and change history with you?
- What real-world action will test the next important assumption?
- What change in your life or work would prove that the partnership helped?
These questions reveal whether you are building an obedient servant, a hidden boss or a correctable partnership.
The most important measure is simple: the purpose of AI partnership is not to increase the role of AI in a person’s life. It is to increase the person’s capacity in real life.