The moment more than one person uses AI on the same job, output quality becomes a function of who typed the request. One person supplies the price list, another does not. One asks for gaps to be flagged, another lets them be filled. The fix is a shared Context Stack per job, agreed by the people who do that job.
Key takeaways
- Variance across staff is a shared-Stack problem, not a training problem.
- Agree one Stack per job type, owned by one person who maintains it.
- The Rules layer is where team agreement matters most, because it encodes your standards.
- Write down what never goes into a general tool before you roll anything out.
- Review the Stack when prices, services or standard answers change, not on a calendar.
Why individual use goes wrong at scale
| What happens | Consequence |
|---|---|
| Each person writes their own instructions | Five versions of the same document |
| Some attach source material, some do not | Some drafts carry invented figures |
| Nobody records what fixed a bad output | The same failure recurs monthly |
| Good prompts live in one person's head | Capability leaves when they do |
| No agreed data boundary | Someone eventually pastes something they should not |
Setting up a shared Stack
One. Pick the job. Quoting, enquiry replies, reporting, onboarding documents. Highest volume first.
Two. Put the people who do that job in a room and agree the Rules layer. This is the part that needs humans, because rules encode standards and standards are a management decision.
Three. Assign one owner. Their job is keeping the Source layer current and adding rules as failures appear.
Four. Store it where everyone works: a shared document, a shared note, or a saved project in your AI tool.
Five. Agree the data boundary in three lines: who may use AI, on what material, and what never goes in. How to set team rules.
The review question
A shared Stack also settles who checks the output. Name that person per job type, not per document, so nothing leaves the business unread. Human approval in workflows.
What changes for the business
Output becomes consistent because the inputs are. New staff inherit a working process instead of a folklore of prompts. And the Stack itself becomes an asset: a written description of how your business produces its documents, which is the same thing you would need for a handover, a hire or a sale.
Build one with your team
Private onsite training builds shared Stacks with the people who do the work, at your workplace, scoped for roughly 4 to 20 people with larger teams split across sessions. Everyone brings the laptop they will use afterwards. Enquire about team training.
For the structure itself, read the full Context Stack, or start from the blank template.
Frequently asked questions
How many Stacks does a team need?
One per job type. Most small businesses run three to five: quoting, enquiry replies, reporting and one or two internal documents.
Who should own a Stack?
The person who does that job most often, not the most technical person. Ownership means keeping the Source layer current and adding rules when something breaks.
How do you get staff to actually use it?
Make it faster than the alternative. A Stack that produces a usable draft in the right shape gets used. One that needs reformatting gets abandoned. Why format matters.
What if staff are already using AI informally?
Collect what they have built. The best rules in most businesses already exist inside someone's personal prompts and have never been written down.
Do you need an AI policy first?
Three lines is enough to start: who, on what material, and what never goes in. A longer policy can follow. Team rules guide.
How often should a shared Stack be reviewed?
When prices, services or standard answers change, and whenever something gets through that should have been flagged. Testing and maintaining workflows.