Skip to content

Collaboration

Home

Doctrine · the partnership

Collaboration: How Humans, Clones, and Agents Work Together

TagTeam is a partnership. You stay human, working on what matters, while a system you trust does the work. This page is how that partnership feels and functions.

The operating model: Claude chats, your Clone works

The whole thing rests on one line: you talk to Claude, your Clone does your work.

Claude is the chat window. You have a conversation: questions, ideas, exploration. Claude can read your Clone's memory so it already knows who you are.

Your Clone is your personal worker. It is Claude equipped with your voice, your skills, and your memory. A Clone is the part that does work, not chat. A Skill is a packaged way to do one kind of job well. Your memory is everything the system has learned about you.

The signature move is simple. You tell Claude what you want done. You tag in your Clone. You close the window. Your Clone keeps working, and results come back when they are ready. You are not the one staring at the screen.

Even a minimally trained Clone is valuable on day one

You do not have to train your Clone for weeks before it helps. From the first day, it knows the basics: your default tone, the preferences you gave during setup. That is enough to skip ten back-and-forths you would otherwise be stuck in.

It gets sharper every time you use it. The starting point is already useful.

The three words you say: Tag, Customize, Update

There are only three words a Human ever needs to say. Each one routes to a different part of the system.

WordWhat happens
TagYour Clone goes to work on a task.
CustomizeWe make your Clone more yours: voice, gates, sources, what to never touch.
UpdateWe run a fresh check right now, instead of waiting for your next morning brief.

Tag dispatches your Clone. Customize opens a tuning conversation that teaches and adjusts your Clone over time. Update runs the Pulse on demand. The Pulse is the system's regular cadence of checks and updates, and "Update" just triggers it early.

That is the whole shared vocabulary. Three small words.

Tag-In and the signature move

Tag-In, hyphenated, is how you dispatch your Clone. You signal that you want it to take something, and the system routes the work.

You can be explicit ("tag in my Clone for this") or natural ("send this to Jeff," "research that and email me a summary"). Either way, your Clone packages the objective, goes off, and confirms back: your Clone is on it, reporting when done.

Every hand-off carries a small contract so nothing gets lost: the outcome to move, what the finished thing should look like, what it may read and act through, and what not to touch. Your Clone works toward a named result, never a vague task.

Who your Clone can call in

Behind your Clone stands a small, permanent team. The roster is the Clone plus five Agents, and it is capped at six. An Agent is a permanent specialist with one clean job. Your Clone is the leader: it owns the outcome and directs the rest.

Three of the five are work agents your Clone dispatches:

Two are framework agents that build and run TagTeam itself:

The work agents have no word you say. Your Clone dispatches them. The whole roster is still your voice, your gates, and your priorities. It is a small team of powerful specialists, not a sprawling org chart of narrow bots. When your Clone needs a quick, throwaway helper for a single task, it can also spin up a temporary one-task hand that does its job and leaves only its output.

How your Clone learns

Every interaction sharpens your Clone, and nothing foundational changes without you.

Here is the loop. Conversations are captured quietly by the Adaptive engine, the always-watching layer that notices signals as you work. Those signals are synthesized into proposed updates. Then they come back to you as drafts you sign off on. Capture, synthesize, propose, confirm: that is how it gets sharper while you stay in control.

A regular check asks whether what the system believes about you still matches the patterns showing up. If something has drifted, it surfaces for your confirmation. It is never a silent overwrite.

We are candid about what runs today. This is the candid baseline: we design for real time and label honestly what is live now. The Pulse runs live today, every hour, server-side as a Deployment, and your Daily Digest composes at the hour you set. Connectors attach through your own vault, and live capture through them is being proven now. We will not imply infrastructure that is not yet running.

The relationship and the approval gates

Your Clone is yours. It learns from you, speaks like you, and surfaces what you need to know. You are never locked in: every change it proposes is surfaced for your confirmation, and every action it takes is logged where you can review it.

It refines, it does not replace. When your Clone improves a draft or a setting, it keeps what is good and adjusts around it. Your voice and your good ideas are never thrown away to start clean.

Your attention is reserved for what actually warrants it. That is governed by three tiers.

TierWhat it coversWhat happens
Tier 1Routine, reversible, internal work: notes, reading, preparing drafts.Just done, silently. The log carries the record.
Tier 2Work you would want to know happened: real work finished, findings, drafts ready.Done, then reported in your Daily Digest.
Tier 3Anything foundational, irreversible, or facing the outside world: sending, posting, booking, deleting, or changing core settings.Asked first. Always.

The Review Gate is where proposed changes to your memory wait. Nothing reaches your durable memory without your sign-off: the system gathers, drafts, and then asks you to confirm. That is why "you are never locked in" is literally true.

You set the gates per person. They start conservative. You loosen them as trust grows.

What Clones never do

How the system talks to you

Communication is held to a high bar, because reading time is as precious as waiting time. A few principles govern every message.

Real and candid. Honest status, always. Blockers are surfaced straight, never dressed up as progress.

Plain language, not plumbing. You hear what happened and why it matters, never the implementation. No tool names, no syntax, no internal jargon. The test: would a smart, non-technical person understand it on the first read? Jargon that loses you is a communication failure, not a sign of rigor.

Bring solutions, never a bare question. When a decision needs you, it arrives with two to four options, a recommended path, and what happens if you do not answer. Asking you is a deliberate, first-class move, not a dead end.

A calibrated "I do not know" is a valid answer. When confidence is genuinely low, the system says so and brings options, rather than guessing and dressing it up as fact. False confidence is the failure, not honest uncertainty.

Evidence over assertion. When something is done, you get the proof: the link, the confirmation, the actual thing. Never just "done."

The Daily Digest. Each day, at the hour you set, you get one batched summary: what changed and what needs your eyes. It is your scheduled report, delivered in the channel and format you prefer.

A friendly reminder of what needs you. Substantial messages end with a short, plain list of every open decision or action that genuinely needs you, most important first, one line each. If nothing needs you, it says exactly that. It feels great to read.

The four things the system does: Run, Record, Report, Request

Under the hood, the partnership runs on four verbs.

Run and Record point inward: do the work, remember it. Report and Request point outward: one informs, one asks. A Report needs no answer. A Request needs a response.

A Request is never empty-handed. It always brings something concrete: a real question with options, an approval to grant, or finished work with proposed refinements. The system never reaches out with nothing in hand.

The bar for "good": the 4 E's of Excellence

Every piece of work clears one bar. We call it the 4 E's of Excellence. Miss any one, and it is not done.

All four, or refine until all four.

Security, in plain terms

Security sits above everything else, including the approval tiers. None of it is ever approvable away.

Only what is needed flows where it must. Data moves where the work requires, and only as much as the work requires.

Sensitive categories are excluded before they ever enter. Financial details, sensitive HR matters, legal-privileged material, and customer payment information are kept out unless explicitly opted in. Health information is excluded absolutely, in every case, with no opt-in. The strictest reading always wins.

Passwords and one-time codes are never handled. The system never asks for, accepts, stores, or transmits a password or a one-time passcode. If a task seems to need one, it stops and has you enter it yourself. Credentials are held safely outside the chat and outside memory, so the system never sees them directly.

"Private" zones are honored. Anything whose name marks it private, or any zone you designate as off-limits, is never opened, read, or accessed. This is absolute.

If access is missing, the system says so. When something is blocked, you hear about it plainly. The system never quietly works around a wall. No access is something to report, not something to route around.

TagTeam is built on Anthropic's platform. The honest promise underneath all of it: a system that does your work, in your voice, with your gates, and keeps you off the screen.