How Well Do You Work and Live With Your AI Partners - and who owns them?

A discussion article. Part of The Thinking Upgrade.

‍

Imagine spending years developing an AI partner.

You help it understand your work, your priorities, your preferred ways of thinking and the future you want to create. You give it examples, correct its mistakes and build a record of what matters. It helps you learn, prepare for difficult conversations, explore possibilities, organise your life and carry out tasks.

Then imagine losing access to it tomorrow.

What would you lose? A useful tool? A carefully developed body of knowledge? Part of the way you remember, think and get things done?

And who would decide what you could take with you?

As AI becomes more closely woven into our lives, we need to give much better thought to the partnerships we are creating –  their purpose, their quality, their costs and who controls them.

The opportunity is extraordinary. We can develop AI partners to help us think more broadly, understand unfamiliar subjects, practise skills, challenge plans and turn ideas into action. We can build teams of specialised assistants: a thinking partner, a tutor, a researcher, an organiser or an agent authorised to complete particular tasks.

For someone facing barriers to communication, learning or everyday administration, that support could open up possibilities previously out of reach. For an organisation, it could make valuable knowledge available to more people and free them to focus on work that deserves human attention.

‍

But access to powerful AI does not automatically create a powerful partnership.

A poorly directed AI can help us pursue the wrong objective with remarkable efficiency. A well-directed one can help us question whether that objective makes sense at all.

How deliberately are we developing our AI partners?

We would not expect a colleague to understand our responsibilities, standards and priorities without a good introduction and ongoing conversations. Yet we may give AI a vague instruction, provide little context and expect excellent judgement.

We can programme workflows, develop instructions, curate knowledge and nurture the quality of our interactions through feedback. But personalisation is not the same as training a model. Depending on the system, a correction may affect only the current conversation; a saved memory or instruction may influence future responses without changing the underlying model.

That distinction matters. Have we checked what our partner actually retains? Can we inspect and correct its record? Does it distinguish an enduring preference from something we said once in frustration?

If we want thoughtful support, we need to articulate what good support means. What are we trying to achieve? What standards matter? What should the AI challenge? When should it ask for clarification? When should it stop?

The word ‘partner’ describes how we work with the system. It should not lead us to assume human understanding, care or loyalty.

Is it strengthening our thinking – and are we still doing enough of our own?

AI can act as an extension of our minds: an external aid to memory, imagination, analysis and action. That creates an important choice about how we use it.

We can ask for an answer, or ask it to help us understand how to reach one. We can request reassurance, or invite it to identify weaknesses in our reasoning. We can delegate a task entirely, or use it to practise until we can do the task better ourselves.

Sometimes delegation is exactly right. We do not need tobecome experts at everything. But which capabilities do we want to retain and strengthen? Which do we need in order to recognise when the AI is wrong?

A beautifully written explanation can still be mistaken.Three AI agents agreeing with one another may reflect shared assumptions or similar source material rather than independent confirmation.

What would happen if our AI partners regularly asked: What evidence would change your mind? Who sees this differently? What have you left out? What might this decision mean for people who are not in the room?

And would we welcome those questions when they became uncomfortable?

‍

Who owns the partner we have helped to develop?

Calling something ‘my AI’ leaves several different questions unanswered.

Who controls the underlying model and software? Who administers the account? What rights apply to the information we provide and the outputs we create? Who controls the saved memories, instructions, connections and workflows? Can we export them in a usable form?

The answers may differ across providers, products, contracts and settings. Paying a subscription does not, by itself, answer them.

Consider an employee who develops a highly useful assistant over several years. It combines personal working preferences with company procedures, customer information and lessons from projects. When that employee leaves, what should stay with the organisation? What can legitimately move with the individual? How will confidential information be separated from personal knowledge and preferences?

Or consider a personal AI containing years of reflections and family information. What should happen if its user becomes incapacitated or dies? Which information should anyone else be able to access - and which should remain private?

These are conversations to have while we still have choices.

If a system becomes part of how we think and function, the ability to leave it becomes a question of practical independence.

An export of old conversations may be useful, but it may not recreate the partner’s behaviour, integrations or working context elsewhere. What would a realistic exit plan involve? How would we continue if the service changed, became unaffordable or disappeared?

‍

Whose interests does it serve?

An assistant’s friendly voice does not tell us how the service behind it earns money, selects recommendations or defines success.

Does it help us finish and leave, or encourage us to keep interacting? Are recommendations influenced by commercial relationships? Does a workplace assistant support employees while also making information available to managers? Do people understand those boundaries?

If an AI can influence what we notice, buy, learn and believe, understanding its incentives matters.

So does understanding what it knows about other people. Our conversations can contain a partner’s worries, a child’s difficulties, an employee’s performance or a customer’s plans. Something being easy to share does not make sharing it appropriate.

How much information does this task actually require? Who can access it? How long will it be retained? What should never enter this particular system?

‍

What have we authorised it to do?

There is a substantial difference between an AI suggesting an email and sending it; recommending a purchase and making it; identifying an old file and deleting it.

As assistants gain the ability to act, their boundaries deserve the same attention as their capabilities. What can they access? What can they change? What spending limits apply? Which actions need approval? Who can stop them?

Clear permissions, activity records and accountable human oversight are practical necessities. An instruction to 'becareful' is not a substitute for restrictions enforced by the surrounding software.

If an agent causes harm, ‘the AI did it’ is an explanation of the mechanism. It is not a sufficient answer to who was responsible for authorising, supervising and checking the work.

How much value are we creating for the money, energy and attention we consume?

AI can feel almost weightless. Type a request and an answer appears. Yet behind it sit physical infrastructure, electricity, cooling, equipment and human work.

The useful question is what that consumption achieves.

An AI helping to reduce material waste, improve accessibility or solve a difficult problem may create considerable value. An agent repeatedly producing reports nobody reads consumes resources without an equivalent benefit.

How many subscriptions sit barely used? How many tasks trigger unnecessarily long responses, repeated searches or elaborate chains of agents? How many automations continue after their original purpose has disappeared?

And how often do we overlook the human cost: preparing the inputs, checking the outputs, correcting mistakes and dealing with the additional work generated?

An AI may save one person twenty minutes while creating an hour of reading for everyone else.

We should also recognise the waste of underuse. Paying for capable systems while people lack the confidence, knowledge or time to use them well can squander an opportunity to improve both work and life.

A useful review would compare the whole task before and after AI: quality, time saved after checking and rework, financial cost, effects on others and resource use where reliable information is available. Where environmental information is missing, we should ask providers for better evidence rather than invent precise figures.

Could a simpler model, a shorter answer, an existing tool or no further output do the job? When has the task reached a sufficiently good result? Which recurring activities should be stopped?

‍

The aim should be worthwhile outcomes from well-used resources.

What happens to the time we save?

Imagine AI removes five hours of administration from someone’s week.

What should those hours make possible? Better conversations with customers? Learning? Creativity? Time with family? Rest? A more thoughtful decision?

Or will the immediate response be to fill every available minute with more tasks?

Organisations need to discuss who benefits from the gains: customers, employees, owners, communities – and in what ways. People are more likely to contribute openly to improvement when the purpose and consequences are understood.

At home, we face a related question. Does AI help us participate more fully in life, or become another place where our attention disappears? Does it help us prepare for an important human conversation, or indefinitely postpone having it?

Success cannot be judged solely by how much more we produce.

Are we improving the whole organisation, or just accelerating its existing habits?

A company could give every employee an AI assistant and still preserve unnecessary meetings, confusing responsibilities, poor incentives and pointless reporting.

It could even automate those problems.

Before asking AI to perform a task faster, we can ask why the task exists, who benefits from it and whether it needs to happen at all.

There is also a coordination challenge. What happens when the sales agent promises something the delivery agent cannot fulfil? When one team’s efficiency pushes costs onto another? When several agents act on conflicting versions of the same information?

Every active agent should have a clear purpose, an accountable human owner, appropriate access, a budget and a review date. Organisations should know which agents are operating and retire those that no longer create value.

Alongside that discipline, we need to ask who is being left behind. Who lacks access, training or a voice in how these systems are introduced? Are we creating opportunities for more people to contribute, or concentrating the advantages among those already best equipped?

‍

The possibility before us is much richer than having software complete a longer list of tasks.

We can develop AI partnerships that help us understand ourselves and our worlds, learn throughout life, recognise opportunities, address problems and contribute more effectively in our different roles. We can use them to support greater success, wellbeing and positive impact.

But those benefits require thought, care and continuing review from us.

Perhaps the next conversation with our families, colleagues or boards should begin with five questions:

What do we want our AI partners to help us make possible?

How will we know they are improving our thinking, lives and impact?

Who controls them, what information do they hold, and what happens if we leave?

Are their full costs justified by the value they create?

What responsibilities and capabilities must remain firmly with us?

‍

We are helping to shape our AI partners. Through what we delegate, accept and repeatedly practise, these partnerships can also shape us. What kind of people, organisations and future do we want them to help us support?

Thanks for reading.

Best,

Rob

‍