What is personal intelligence?

Personal intelligence is the capacity to reason about personality and use personal information to guide decisions, plans, and goals. The term was introduced by the American psychologist John D. Mayer in a 2008 paper in Imagination, Cognition and Personality. Mayer's framework identifies four core capacities, all of which are now being implemented in software for the first time. moccet is one of the systems built around them.

This essay explains what personal intelligence is, why the term matters more than the looser phrase personal AI, and what changes when the capacity belongs to a machine that runs continuously alongside a person's life rather than to a small number of trusted humans whose attention has historically been priced as a luxury.

What are the four capacities of personal intelligence?

Mayer's 2008 model identifies four components, validated across two decades of empirical work including the Test of Personal Intelligence developed by Mayer, Panter, and Caruso in the Journal of Personality Assessment (2012).

The first capacity is recognising personally relevant information. A high-PI person notices the signal that matters in the noise of an ordinary day. The off-handed comment that contains the real concern. The pattern in someone's behaviour that suggests something has changed. The detail in a calendar that explains why a week has felt unmanageable. Most useful judgement begins with recognition of this kind.

The second capacity is forming accurate models of personality, both one's own and others'. A model is more than a list of facts. A model is a structured representation that supports inference. Knowing that a colleague communicates indirectly, prefers detail over summary, and trusts data more than narrative is not a list. That information is a working model that lets a person predict how the colleague will respond to a draft before sending it.

The third capacity is using those models to guide choices. A person who has built accurate models of themselves and the people around them makes different decisions than a person who has not, even when the available information is identical.

The fourth capacity is organising goals, plans, and life stories into something coherent enough to act on. Personal intelligence operates across time. The capacity holds a working sense of what a person is trying to do across longer arcs and uses that sense to evaluate what should and should not occupy the next hour.

These four capacities define the category. A machine personal intelligence implements them in software. Recognition becomes a continuous classifier reading across connected sources. Modelling becomes a structured representation of the user, updated as life changes. Guidance becomes the system's actions, taken with confirmation. Goal organisation becomes long-term memory holding projects, commitments, and relationships in coherent form.

Why use the term personal intelligence rather than personal AI?

The phrase personal AI is doing rough work in the current market. Different products use the phrase to mean different things. Some mean a chatbot with memory. Some mean a bounded agent that completes single tasks. Some mean an ambient system that runs across a person's life. The looseness has made it harder for users to tell which products do what.

Personal intelligence draws the right line. The term carries seventeen years of empirical scaffolding from Mayer and his collaborators. Personal intelligence refers to a specific cognitive capacity with measurable components. The four capacities together describe a system that does something fundamentally different from a chat with a memory feature.

A chatbot with a memory feature satisfies, at most, a thin version of the second capacity. The system stores discrete facts the user has shared and retrieves them in future conversations. The facts sit on a list. A list is not a model. The system cannot reason across the facts. The system cannot tell that the project mentioned last Tuesday is the project that explains the late nights mentioned this Tuesday.

A bounded agent that completes a research task or writes code can satisfy a different thin version of the third capacity, in a single domain, for the duration of a single goal. A bounded agent does not run continuously. A bounded agent does not maintain a representation of the user across the unbounded surface of a life.

A genuine personal intelligence, machine or human, satisfies all four capacities simultaneously and continuously. Recognition runs across every connected source. The model is structured, updated, and queryable. Guidance is acted on. Goal organisation persists. The work is integrated rather than parcelled out across separate features.

Integration is the distinguishing technical property of the category. Integration is also the hardest engineering problem in consumer AI today. The companies investing in the architecture rather than rebranding existing chat features are the companies building what comes after the chatbot.

What architecture does a personal intelligence require?

A chat product is built around the conversation. The user opens the application, types a request, receives a response, and leaves. Whatever continuity exists between sessions is supplied either by the user repeating themselves or by a memory feature that retrieves stored facts and inserts them into the next prompt. The architecture has produced ChatGPT, Claude, and Gemini in their conversational modes. These products are excellent at what they were built for. They are not built for the work of running a life.

A personal intelligence is built around the model of the user. The model is the centre. Interfaces, including a chat where one exists, are surfaces around it. The architecture has three properties that the chat architecture cannot easily acquire by adding features.

A personal intelligence is continuous. The system runs without being summoned. Most of the system's work is invisible to the user, because most of the work of running a life is routine and the system handles routine quietly.

A personal intelligence is contextual. The model of the user is updated continuously from connected sources. Every interaction, every action, every decision the system makes draws on the same shared context. The system that drafts an email knows what is on the calendar, knows what was said in the prior thread, knows the relationship to the recipient, knows what the user has committed to for the rest of the week.

A personal intelligence is selective. Most of what the system notices does not warrant the user's attention. A small fraction warrants quiet action. A smaller fraction warrants surfacing. The system that makes these classifications well is the system the user trusts. The system that gets them wrong is the system the user uninstalls.

This is what moccet is being built to be. The architecture is the consequence of taking Mayer's framework seriously rather than treating personal intelligence as a marketing label for a chatbot with extra features. The architectural pattern that supports continuous, contextual, selective operation is the orchestrator-worker pattern that has become the consensus design for modern multi-agent systems.

Who is personal intelligence for?

Personal intelligence delivers something that has historically been available only to a small number of people. The capacity delivers the kind of attention that comes from another mind that knows the whole of one's life and acts on it.

A great executive assistant builds personal intelligence about a single principal across years of working together. The assistant learns the patterns, holds the commitments, maintains the relationships, drafts in the principal's voice, anticipates what is coming, handles quietly what should not need to reach the principal. A long-tenured chief of staff does the same at greater scope. A trusted advisor or therapist accumulates, over decades, an unusually deep model of a single person and uses the model to guide better decisions.

The arrangement scales badly. The number of people who can afford this kind of attention has always been small. The number of practitioners is bounded by the supply of skilled humans willing to do the work. Personal intelligence has, until now, been a service for the small fraction of people who could pay for it directly.

A machine personal intelligence offers a version of the same capacity at scale that approaches the marginal cost of compute. The system does not sleep. The system does not need to be told what happened last week. The model the system maintains is a permanent feature of the substrate, updating in real time across every connected source. The kind of attention previously available only to a small number of people becomes available to anyone willing to delegate it, provided the engineering is done well enough to earn the trust.

The qualifier matters. A poorly built personal intelligence is worse than no personal intelligence. The depth of access required to be useful is the same depth of access that creates catastrophic risk if the system is built without the right architecture. The privacy components, the action sandbox, the confirmation discipline, the audit logs, the right to revoke and delete are not optional features. They are the conditions under which the technology is usable in the first place. The full privacy architecture of a personal intelligence is what separates a system the user has chosen from a surveillance product wearing the costume of one.

What does using a personal intelligence change?

A user who works with a personal intelligence experiences something different from using a chat AI. The difference shows up in the rhythm of the week rather than in any single interaction.

A chat product is something the user reaches for when they have a task. The interaction is bounded. Between tasks, the system does nothing. The user remains the centre of the work and the system is an accessory.

A personal intelligence is something the user notices is on. Most of the work happens without the user looking at it. The dinner that needed to be booked is booked, with a confirmation drafted and sent in the user's voice. The meeting that needed to move has moved, with a note in the user's tone explaining the change. The morning that needed to be protected is held, because the system saw the recovery score, the late nights, the deadline, and made the right call about what tomorrow needed to be. The user reads what happened, occasionally edits, mostly approves, and returns to the work that needed them.

The shift is from doing the integration work yourself to having a system that does the integration. The integration work is what has not gotten easier with the past three years of AI tools. The Boston Consulting Group study published in Harvard Business Review in March 2026 found that knowledge workers using four or more AI tools were less productive than workers using two, because the cognitive cost of integrating across the tools exceeded the per-tool benefits. The user was the bottleneck. A personal intelligence is the architecture that addresses the bottleneck directly, by being the integration rather than asking the user to perform it.

What products are being built now?

Personal intelligence is in its first full year of being technically possible at consumer scale. The category is small. Most of what is sold under the personal AI label is the older shape of product with new marketing. Genuine personal intelligence systems are rare, in beta, or limited to early users. moccet is one of them.

This is normal for a new category. The PC was a curiosity for years before it became an institution. The smartphone was a niche product before it became the operating system of daily life. The first generation of any new category is small. The architectural commitments made in this generation will determine which products are still in use a decade from now.

The companies that take Mayer's framework seriously, that build the four capacities into the substrate of their systems, and that pair the depth of access required with the privacy architecture required to make depth safe, will produce systems that feel quietly indispensable to the people who use them. The companies that ship list-based memory features and call them personal AI will continue to compete in a market that loses interest in chatbots faster than they can add features to them.

For knowledge workers whose lives have outgrown what a calendar and to-do list can hold, the question is not whether personal intelligence will become a real category. Personal intelligence already is a real category. The question is when the engineering will be ready to deliver it at the depth and reliability that justifies handing a system this much of one's life. moccet is being built to be ready.