Enterprise and personal ANC: one architecture for organizing AI work around goals

My idea of a digital counterpart comes from ANC: an AI Native Company. In Lawted's discussion of ANC, I especially liked the idea of building the trunk so different applications and capabilities can grow from it.

As I think further, a digital counterpart increasingly looks like a personal version of ANC.

Enterprises and individuals operate at different scales but share the same question: how can goals, context, working methods and available capabilities be organized so AI understands what to do and keeps making progress?

To understand this, first consider what people currently do.

Today, people organize AI's work

Consider a recruiting project. AI can analyze roles, organize resumes, compare candidates and draft communication. Yet someone still decides which materials to analyze first, what background matters, which requirements take priority, who receives the result and what happens after feedback.

Much of this information is implicit. A client says industry experience matters but may care more about sales-model fit. Only the lead may know whether to wait for a candidate or change direction. A rejected recommendation needs interpretation: was the issue the candidate, criteria or our understanding of the client?

People carry goals, background, judgement and progress, while connecting different capabilities.

Even as every AI tool improves, coordination can remain the bottleneck. Analysis may be quick; choosing what to analyze, explaining context, checking results and arranging follow-up still consume attention.

That is my starting point for ANC: gradually organize information that people currently carry and connect into an environment AI can use.

Enterprise ANC: make goals and working methods available to AI

I understand enterprise ANC as an architecture organizing information, capabilities and collaboration around company goals.

An agent needs to know the objective, why a task matters, business facts, working methods, available tools, necessary participants and current progress. Together, these form the context for action.

In recruiting, imagine an agent reading the client background, role criteria and candidate records after receiving a hiring goal. It determines what can start and what information is missing. If a key requirement is contradictory, it asks the lead to confirm with the client and explains which decisions depend on the answer.

While waiting, it records whom it is waiting for and what it needs, and continues unaffected work. When confirmation returns, it updates its understanding, reassesses candidates and arranges follow-up.

Notifying someone is only one step. The whole process includes asking the question, retaining waiting state, understanding the answer and continuing.

People supply real-world information, necessary authorization and key judgement. Agents organize and execute within their capabilities.

This explains the trunk metaphor. Role analysis, candidate evaluation and interview reports are leaves. Shared client understanding, criteria, task state, permissions and feedback mechanisms are the trunk that lets them work together.

New capabilities can reuse existing context; useful experience from completed work supports later work.

Personal ANC: apply the same architecture to one person

People have goals, experience, principles, projects and methods. These are often scattered across their minds, documents, conversations and tools. Each AI interaction requires selecting materials, reexplaining context and specifying the next action.

Personal ANC gradually organizes this into persistent personal context that can be maintained and used by different AI systems. That is the role of a digital counterpart.

Someone who wants to write consistently has much more context than “write me an article”: why they write, what readers should understand, which ideas come from experience, which expressions are outdated, why previous drafts were rejected and what is underway.

With appropriately organized context, an agent can judge whether to draft or gather another example, and whether dissatisfaction calls for changing expression or reconsidering the argument. Understanding should continue across models and sessions.

Both share one organizing logic: provide context around goals, organize capabilities for action and revise understanding from results.

Enterprises additionally handle shared responsibilities, permissions and conflicting goals. Personal systems need sustained understanding of one person. These differences affect design but not the underlying logic.

Relationships matter as much as stored information

Externalizing information can sound like saving documents and chats. Storage is only one step.

Client feedback needs its project, the judgement it corrects and its scope. Personal experience needs the conditions under which it arose and whether it remains valid. A task needs its goal, why it paused and what allows it to continue.

These relationships determine how AI uses information. “This client values this experience” and “prioritize this experience for every role” produce different actions. Losing scope can turn correct feedback into a wrong general rule.

The architecture needs to answer: what is this information, why does it matter, when does it apply, how does it relate to the current goal and what changes with new feedback?

Knowledge, skills, task records and context serve different purposes. Their relationships, timing of delivery to AI and updates are central to information organization.

Usable information also needs tools, authorization and execution mechanisms before judgement becomes action. Effectiveness must ultimately be tested in actual work.

Context develops through collaboration

One interview or comprehensive profile cannot finish this work. We often know more than we can articulate.

A person may not explain how they assess candidates but can identify a poor recommendation. They may struggle to define good writing beforehand, yet explain why a specific draft fits or misses the point. Choices and corrections expose implicit judgement.

Context therefore needs an ongoing process: work from existing understanding, find gaps through real tasks, incorporate human feedback, organize it, assess its scope and check its usefulness later.

One correction is not automatically a general method; one success does not prove understanding. The framework must allow updates, exceptions and withdrawal of wrong conclusions.

Agents should proactively help organize and maintain context. Requiring users to summarize every lesson, fill out background and update workflows makes externalization another burden.

Why this architecture matters more as AI improves

My work on a digital counterpart assumes AI will be able to do more. Steps requiring people today may become agent work later. But if goals, background and relationships remain only in our minds, every capability improvement still needs human reorganization to create value.

Human–AI responsibilities should evolve with capabilities and authorization. Context should accumulate independently of one model so new capabilities can continue existing work.

Model upgrades improve general capability, but why a company operates as it does, why someone pursues a goal and why previous judgements changed still come from real activities.

Enterprise ANC organizes a company's goals, context and collaboration. Personal ANC organizes a person's goals, context and actions. Both provide foundations for sustained understanding and progress.

For me, a digital counterpart explores personal ANC: give experience and goals a form AI can understand and use, then refine it through long-term collaboration.

This method is still evolving. Its test is whether AI makes more suitable decisions, continues work and reduces repeated explanation and step-by-step direction.