Collaboration
This is how Humans, Clones, and Agents actually work together. The whole point is simple: you get your time back, off the screen, while the work still gets done.
The operating model: Claude chats, your Clone works
There are two surfaces, and keeping them straight makes everything click.
Claude is the chat window. You talk to Claude the way you do now: questions, ideas, thinking out loud. Claude can read your Clone's memory, the saved picture of who you are, so it already knows you.
Your Clone is your personal worker. A Clone is Claude equipped with your voice, your skills, and your memory, running on your behalf. A Skill is a packaged set of instructions that gives a worker a real capability. Your Clone does work, not chat. You hand it a job, it goes off and executes, and it reports back when results are ready.
Behind your Clone stand three Agents, autonomous workers that each own one job: one gets you set up, one builds new parts of the system, and one keeps everything sharp day to day. Your Clone calls these in as needed. You never have to think about them.
Tag-In: how you hand off the work
Tag-In is the move where you tag in your Clone to take something off your plate. You stay in the chat with Claude. When you signal that you want work done, your Clone gets dispatched and goes to work.
The signature TagTeam move, in one line:
You tell Claude what you want done. You tag in your Clone. You close the window. Results come back when they are ready.
You can tag in plainly ("tag in my Clone for this") or just describe the work ("research this and email me a summary"). Either way, your Clone confirms it is on it, does the job in your voice, and reports back through whatever channel you prefer. You are not the one staring at the screen.
Valuable from day one
You do not need to train your Clone for weeks before it helps. Even a minimally trained Clone is useful from the first day. It already knows your basics: your default tone, your top priorities from setup. That is enough to handle a lot of back-and-forth for you, so you skip ten rounds you would otherwise be stuck in.
How your Clone learns
Your Clone gets sharper every time you use it. The mechanism is steady and simple:
- Signals are captured. Conversations and observations land in a working layer where new signals collect as they come in.
- Signals are synthesized. The system reads those signals and looks for real patterns in how you actually work.
- Changes are proposed back to you. Anything meaningful about who you are arrives as a draft for your yes, never a silent change.
Candid baseline (so we never oversell it): this capture and the learning cycle run daily today, triggered, and are built to go continuous when live connectors and scheduling arrive. The engine is already designed for always-on. We will not call it real-time or always-on until it truly is.
Nothing foundational about you changes without your sign-off. Background capture is free and reversible. Every proposed change to your profile, your goals, or your settled truths waits at a checkpoint until you approve it.
The relationship, in plain terms
Your Clone is yours. It learns from you, it speaks like you, and it surfaces what you need to know. You are never locked in. Every change it proposes is shown to you first. Every action it takes is logged where you can read, review, or refine it any time.
It also refines rather than replaces. When your Clone improves a draft, a setting, or a profile, it keeps what is good and sharpens around it. Your voice and your good ideas are never thrown out to start clean.
What needs your approval
Your attention is reserved for what actually warrants it. Three tiers govern this. The rule of thumb: small reversible things just happen; new or fundamental things ask first.
| Tier | What it covers | What happens |
|---|---|---|
| Tier 1 | Routine, reversible background work: notes, captures, logs, reading, preparing drafts | Just done, silently |
| Tier 2 | Background work you would want to know happened: real work finished, notable findings, drafts ready | Done, then reported to you |
| Tier 3 | Anything new, lasting, or sent outside: your profile, your goals, sending or booking or deleting, anything touching money or other people's time | Asks you first |
You also set your own gates: what your Clone can act on alone, what it must draft for your review, and what it must always escalate and never do on its own. The default is conservative. You loosen it as trust grows.
What Clones never do
- Chat. (Claude chats; Clones work.)
- Speak for anyone else. Every Clone belongs to one Human.
- Override your approval gates.
- Touch family or personal-life matters without your explicit opt-in.
- Make an unrecoverable decision without confirming with you.
- Send, post, change anything lasting, or commit anything outward-facing without your approval.
How your Clone communicates
The communication is built to respect your time and your trust.
- Real and candid, always. Honest status over polish. If something is blocked or not done, you hear it straight, never dressed up as progress.
- Plain language, not plumbing. You hear what happened and why it matters, in human terms. No tool names, no technical jargon. If a detail would not change your decision, it gets cut or translated into plain outcome language.
- Solutions, never a bare question. When a real decision needs you, your Clone brings two to four options, a recommended path, and what happens if you say nothing. You are never handed a naked "what should I do?"
- Evidence over assertion. "Done" comes with the proof: the link, the draft, the confirmation. Never just a claim.
- The Daily Digest. Each morning you get one digest: what changed, what got handled, and what needs your eyes. The Daily Digest is how routine work reaches you in a batch, instead of pinging you all day.
- The friendly reminder. Substantive updates end with a short, plain list of any actions that genuinely need you, most important first, one line each. If nothing needs you, it says exactly that. Nothing low-priority ever clutters it; that rides the digest instead.
When an engineering or operational snag comes up, your Clone does not bog you down with it. It explains briefly, picks the path that ships your outcome, logs the issue, and moves on. Taste, brand, voice, strategy, money, and relationships come to you. Plumbing does not.
Security, in plain terms
Security is foundational here, not bolted on later.
Only what is needed flows where it must. Data moves only where it has to, and only as much as it has to.
Sensitive categories are excluded. Certain things are filtered out before they ever enter the system, not cleaned up afterward. At the company level that means financial details, sensitive HR matters, legal-privileged material, and customer payment information. At your personal level it means personal email categories, anything marked confidential, and family threads. Health information is excluded absolutely, in every case, with no opt-in. Meeting transcripts get the strictest handling of all, because they are the most sensitive thing the system ever reads.
Secrets stay off the system. Passwords, keys, and tokens never live in chat, in memory, or with the workers. They stay locked down on the host, and the workers never see them directly. We will never ask you to paste a credential into chat.
If access is missing, you are told plainly. When the system cannot reach something, it says so and brings you options. It never quietly works around a missing permission. No access is something you hear about, not something the system improvises past.