Shared AI chats: how a team can work with AI agents together, safely

At most workplaces, AI is a solo activity. Each person works in their own window, copies the good bits into an email, and the next person starts again from nothing. The thinking stays locked in private chats. There is a better way: let the team work with the same agent, in the same chat, at the same time. The trick is doing it safely.

Why solo AI hits a ceiling

When everyone uses AI alone, three things go wrong.

• Context gets copied around. You paste the brief into your chat, a colleague pastes a slightly different version into theirs, and the two agents give two different answers.
• Know-how stays in one head. The person who worked out how to get a great tender response from AI keeps that knowledge in their own chat history.
• Hand-overs are lossy. “Here’s what the AI said” in an email drops the reasoning, the sources and the corrections that got you there.

A shared chat fixes all three. Everyone sees the same conversation, works with the same agents, and picks up from the same place. The agent becomes a member of the team, not one person’s private tool.

Start private, share when it is useful

You do not need to plan a shared chat from the start. Work something out on your own first, with all the false starts that involves. When it would help someone else, share it.

Everyone you add sees the whole chat, earlier messages included, and works with the same agents as you. Nobody has to re-explain anything.

There are two kinds of shared chat worth knowing:

• Group chats are with the people you pick. Think of a group chat on your phone.
• Team chats belong to one team, such as Marketing or Sales. Everyone in the team is in them, people who join the team later see them too, and they start with the team’s instructions, files and logins.

The rule that makes it safe

Here is the question to ask about any shared AI tool: whose permissions does the agent use?

The safe answer is that the agent acts with the rights of whoever sent each message. Every request is checked against the person who sent it, and the agent only does what that person is allowed to do.

That one rule makes a lot of things possible:

• You can bring in your own knowledge. Ask the agent to look through your private chats and pull out what you worked out last week. Only the answer you asked for is posted into the shared chat. Nothing else from your private chats is shown.
• Your colleague cannot follow up for you. If someone else in the same chat asks for more detail from your private chats, the agent says no. It cannot read them on their behalf.
• Files work the same way. The agent can answer you from a folder only you can open, without copying those files into the shared chat.
• Logins stay where they belong. Your private credentials only work in a shared chat if you share them with everyone in it.

Try saying:

Find what I worked out about the new pricing in my private chats last week, and summarise it here for the team.

Without a rule like this, a shared agent would be a leak waiting to happen. With it, a team can do real work together.

Read back before you add people

Sharing a chat shares all of it. That might include a draft with a client’s personal details, a frank comment about a colleague, or a figure that is not public yet. Read it back before you invite anyone.

Try saying:

Before I share this chat with the team, list anything in it that looks personal, confidential or not meant for a wider audience.

If the chat has too much in it, start a fresh chat for the team and ask the agent to bring across only what they need.

Say who does what

With several people and the agents in one chat, be as clear as you would be in a meeting. Name people for each part.

Try saying:

Agent, pull last quarter’s sales figures into a table. Sam, can you check the figures it finds against the finance report? Priya, once Sam is happy, draft the summary for the board.

A few other requests help a busy chat stay organised:

Try saying:

List what each person in this chat agreed to do, by name, with dates.

Try saying:

Summarise this chat so far for Jordan, who just joined.

When the agent needs a decision, it should be clear who it is waiting on. In a good shared tool, a question goes to the person who asked for the work, and only they can answer it.

Keep the main chat clean with side chats

A shared chat gets messy fast if everyone asks their own small questions in it. Side chats solve that.

A side chat is a private branch of the conversation. Its agent knows everything said so far, but the main chat stays exactly where it was, and nobody else is notified or distracted. Ask what step 4 actually means, check a figure, or try a rough idea, then go back to the main chat from the same spot.

If the side chat turns up something everyone should know, ask it to share.

Try saying:

That’s useful. Share this with the main chat.

See where a colleague is going, as they type

In a busy team chat, waiting for someone to finish a long message slows everything down. Live drafts let others see your words as you type them, not just “Sam is typing”. A team can build a message together and react to an idea before it is sent.

It should always be your choice. Good tools show clearly when live drafts are on, let you switch them off for yourself everywhere, and never save the drafts.

Say it once with team instructions

Every team has knowledge that lives in people’s heads: the main clients and what they care about, the house style, where the templates are, the rules everyone follows. Put that into team instructions once, and every team chat starts already knowing it.

Good things to include:

• Who your main clients are and what they care about.
• House style: tone, spelling, how documents are laid out.
• Where things live: “Price lists are in the Sales team folder.”
• The team’s rules: “Quotes over $10,000 need a manager’s sign-off.”

Keep it short and true, and check it every few months so it does not go out of date.

Try saying:

Read the price list in the Sales team folder and draft team instructions for our quoting rules and tone, ready to paste into Team settings.

Turn your best prompts into team skills

Most teams have one or two people who are really good with AI. Skills let everyone else get the same results. A skill is a set of instructions, plus documents if you like, for one kind of job: a report format, a review checklist, a writing voice. The agent uses it whenever a request matches.

Good candidates are prompts you have refined over time. The tender review checklist. The way you write board papers. The client onboarding email.

Try saying:

Write the checklist in this chat up as a skill: a name, when to use it, and the instructions, so I can add it to the Marketing team’s skills.

A simple way to begin

1. Pick one piece of work your team already does together, such as a weekly report or a proposal.
2. Start it in a private chat, then share it with the two or three people involved.
3. Name who does what, and use side chats for the small questions.
4. When the same background comes up twice, move it into team instructions.
5. When the same job comes up three times, turn it into a skill.

How Interweave OS does it

In Interweave OS, every chat can become a group chat: press Share and add people or a whole team, and everyone carries on with the same agents and the whole history. Team chats start with the team’s instructions, files, credentials and skills. The agent acts with the rights of whoever sent each message, so private chats, files and logins stay private, and a colleague can never ask it to read yours. Side chats open beside any chat for the quick question, and live drafts can be turned on for a chat or a whole team, with your own Never setting always winning.