Talk me through what a good digital teammate should and should not do.

A Digital Teammate should collaborate with a human, not replace one.

Over the past few weeks I’ve been asking ChatGPT to generate a series of interview-style questions designed to make me think more deeply about topics I care about. The exercise was intended as self-reflection, but I found that many of my answers naturally evolved into short essays. Rather than leave them buried in my notes, I’ve decided to publish them as well.

The Question:

Talk me through what a good digital teammate should and should not do.

My answer:

A good Digital Teammate is closer to an operational partner, it helps turn intent into finished work by holding context, preparing material, and keeping the workflow moving. It can take on bounded tasks, but it should not decide what matters or what I believe. I still have to choose the direction, make the calls that need judgement, and remain accountable for the result.

This is how I think knowledge work is starting to evolve: As AI capabilities improve, the human role shifts away from carrying out every task manually and towards deciding what should happen next. The AI takes on more of the operational load, especially where the work is easier to check than to do from scratch.

In practice, that collaboration works best as a sequence rather than a single prompt. A good Digital Teammate might first brief me on the evidence, then ask the questions that force me to make my own view explicit. Only after that should it help turn the thinking into a draft I can challenge and revise.

That is very different from asking an AI to “produce a finished strategy document”. In one case, the AI is helping me think and execute; in the other, it may be producing something that sounds like a view before I have actually formed one.

A Digital Teammate needs context and boundaries

A Digital Teammate is only as effective as the context it has available and the rules within which it operates. It needs enough organisational knowledge to understand the work, and clear enough boundaries to know what it is allowed to do. Without that, AI becomes inconsistent and unreliable. With it, the system can take on bounded work without losing sight of the role it is supposed to play.

In my opinion, the surrounding harness matters as much as the model itself. A capable model may still be unreliable if it lacks durable memory, permissions, and workflow rules. The system needs to know its role, what state it should preserve, and where human approval is required. Otherwise it may confidently do work it should have stopped or checked first.

The human role becomes more important, not less

By 2026, I think a well-designed Digital Teammate can already handle a large share of the manual cognitive work around many knowledge-work tasks. It can gather the material, prepare the first version of the work, and carry routine administration between sessions. That does not mean it can handle the whole job, or that it should be trusted equally in every setting. It means the scarce human contribution moves towards deciding what the work is for and whether the result is good enough to stand behind.

I increasingly see myself less as someone who writes every document from scratch and more as someone who decides what needs to be written. The AI can help challenge the draft and spot what is missing, but it is still my job to decide which gaps matter. I also have to decide what trade-offs I am willing to make, and whether the final answer represents my judgement.

A good Digital Teammate should not become an opinion generator

A Digital Teammate should not decide what I believe. It should help me work out what I believe by giving me something to test, question, and refine. A Digital Teammate can help operationalise my intent, but it should not manufacture the intent itself.

The worst use of AI is asking it to produce work that represents views I have never actually formed. The resulting document may be articulate, persuasive, and well-structured, but it is not genuinely mine. The problem is not only that the writing has been delegated; it is that the thinking may have been delegated without anyone noticing.

The best Digital Teammates push in the other direction. They expose weak assumptions and help me notice when my reasoning is thinner than I thought. Part of their value is that they improve the quality of the questions being asked, not only the polish of the answers.

The interaction should be iterative, not transactional

In practice, I think the relationship works best when it is iterative rather than transactional. I do not want a Digital Teammate that waits passively for isolated prompts and then returns polished prose as if the job were complete. I want one that works through a sequence of collaboration steps that strengthen the thinking as well as the output.

That means the AI should not treat the first plausible answer as the endpoint. It should be able to brief and draft, but also to notice when continuity is missing or an assumption needs testing. The human stays active through the whole exchange, because polished wording does not mean the answer is right.

Good Digital Teammates improve the work before they accelerate it

A Digital Teammate is not simply a productivity tool. If all it does is help people produce more output, then it may just help organisations create low-quality material more quickly. The more interesting possibility is that it improves the quality of the work before it improves the speed of it.

That means challenging weak reasoning and surfacing what is missing, rather than making a half-formed argument sound more confident. It also means preserving the context needed for thoughtful work to compound over time.

The ideal outcome is not just faster execution. It is better judgement expressed more clearly and carried through more consistently.

This is increasingly how I work myself

This is not a hypothetical model for me. It is increasingly how I use my own Digital Teammate.

One ordinary example is my Daily Note workflow. The system can open the day, carry forward unfinished tasks, and preserve useful context from yesterday. That saves time, but the more important test is whether it keeps the day honest. If something is blocked, it should say why. If a task is only deferred, it should not dress that up as a blocker. If I ask it to prepare a piece of writing, it should preserve the difference between my judgement and its draft.

That is the point at which the system becomes useful and risky at the same time. The more it understands my work and previous decisions, the more helpful it can be. The same context would become dangerous if the human role became vague. A Digital Teammate can help me think better, but it should not silently think on my behalf.

The goal is partnership, not replacement

Ultimately, I do not think the goal of a Digital Teammate is to become an autonomous replacement for a knowledge worker, but rather an extension of one.

The human remains responsible for direction and judgement, including the ethical choices that sit behind the work. The Digital Teammate can handle much of the operational execution, provided its authority stays bounded and its work remains open to challenge.

That is the partnership I want: not a system that replaces the human voice, but one that gives the human more room to think clearly and produce work they can still honestly stand behind.

---

My Digital Teammate is part of my SecondBrain system. I have published the reusable parts as the SecondBrain Framework: the structure, operating contracts, templates, and tools without my private vault content.

---

Generative AI is a powerful tool. I openly use it to conduct research, challenge my thinking, generate drafts, and review my work. AI never owns what I think, or my personal judgement. Everything I publish is a reflection of my own thoughts and opinions.