Why your team's AI work stays in one person's chat, and how to fix it

You pay for AI for the whole team. Two people get great results. Everyone else gets so-so answers, or stopped trying.

The good work already exists. Someone on your team wrote the perfect refund reply. Someone built a prompt that turns a call into a clean client report. It works. It lives in their chat history. Nobody else can use it.

Below: why that happens, and what you can do about it this week.

The good work is already there, in one account

Look at how most small teams start with AI. Each person signs up, or gets a seat. Each person opens their own chats. Over a few months, a couple of them get good at it. They learn what to paste in, what to ask, and how to check the answer.

All of that sits in one place: their account. Their saved prompts. Their custom GPT. Their project in Claude. Their notes on what works.

One team lead put it plainly on Reddit:

"every seat is a silo. One person builds great prompts and we pay for 11 other people to not have them. Someone leaves, their setup walks out the door and the bill stays the same."

You are not short of AI. You are short of a shared place for what people do with it.

Why it stays stuck

People don't refuse to share. The tools make it hard, in a few ways.

Chats are personal by design. A chat history belongs to the person who typed it. There is no "send this to the team" button that keeps working when the prompt changes.

Team plans do not carry the work over. When people move from a personal plan to a team plan, their setup often stays behind. One ChatGPT user upgrading to Team wrote: "As well, your Custom GPTs do not transfer over to Teams." Another said: "It's easier to just re-ask and start anew on the Teams account."

Some setups only live on one computer. An admin of a Claude Team account wrote that in Cowork, people's "changes and projects only live on their local machine and cannot be shared to the rest of the team."

People also use their own tools. A large study by the University of Melbourne with KPMG, with 32,352 employees, found: "The majority (70%) say they use publicly available AI tools that are free to use." Only 42% use "AI tools provided by my employer." (KPMG, 2025) If the work happens in a personal tool, it stays in a personal tool.

And good work spreads slowly even when tools allow it. Microsoft's 2026 Work Trend Index looked at what worked and what failed with AI at work. Its finding: "In many organizations surveyed, those signals stay local or spread slowly." (Microsoft, 2026) Microsoft sells AI tools, so read that with care. But it matches what owners say.

What it costs you

• You pay for ten seats and get two people's results.
• Everyone writes their own version. The refund rule ends up in eight prompts, each a bit different. Someone fixes one. The other seven are now wrong.
• New people start from zero. The new hire asks the AI about a client and gets a generic answer.
• Know-how leaves with people. When your best person goes on holiday, or leaves, their way of doing it goes too.

None of this is on an invoice. You see it as slow mornings and uneven work.

How to fix it, step by step

You can start today with no new tool. This plan suits a team of 5 to 50.

1. List the five jobs AI helps with most. Ask the team. Think refund replies, client reports, product descriptions, proposal drafts, weekly updates. Keep it to five.
2. Find who does each one best. Usually one person per job already has a good prompt. Ask them to share it, plus the steps they follow around it.
3. Write each one down as a playbook. One page each. What the job is. What to give the AI. The prompt itself. How to check the result. Plain words, no jargon.
4. Put them in one shared place. A shared folder, a doc, a wiki. Anywhere the whole team can open. Not in one person's account.
5. Give each playbook an owner. One person keeps it current. When the process changes, they update the one copy. Everyone uses that copy.
6. Make it the default. In your next team meeting, show one playbook in action. Ask people to start from the shared version, not their own.

This works, up to a point. People still have to find the doc, copy the prompt and paste it into their AI. The doc goes stale when nobody owns it. And the AI never sees the playbook unless someone pastes it in.

Where Homla fits

Homla is one shared home for how your team uses AI. It does the steps above inside the AI your team already uses.

• Someone types "save this as a skill" in Claude or Codex. It goes into the team library, not their history. See Skills.
• Anyone on the team asks their AI for the refund reply. Their AI uses the team's playbook.
• Fix a playbook once and everyone gets the fix. Every version is kept, so nothing is lost.
• Finance playbooks stay with finance. Client playbooks stay with the client team.
• When someone leaves, their skills stay. The team keeps using them.

It works in Claude, ChatGPT, Cursor and other AI apps. Your people keep the AI they like. Client details and decisions can live in shared Memory too, so nobody explains them again.

Start with one playbook this week

Pick the one job your team does most with AI. Find the person who does it best. Write down how they do it, and share it with everyone.

That one step moves the work out of one person's chat. Then do the next job, and the next.