If you use AI at work, you carry a growing pile of context around with you. Memory entries in a chat tool. A folder of prompts. Project notes you paste in at the start of a session. An instructions file you keep tweaking. Over time that pile turns into a knowledge base, whether you planned one or not.
When I look at what sits in mine, I see three different kinds of information. Each has a different owner, a different lifespan, and different rules about who should see it. Most tools dump all three into one bucket. Leaders should split them apart before the bucket gets big.
Layer 1: Personal information
This layer holds what the AI knows about you as a person. Your name, your city, how you like answers formatted, your family, your health constraints, your dislike of long emails.
A hypothetical example: a product manager tells her assistant she has two kids, coaches soccer on Tuesdays, prefers bullet summaries, and writes in British English.
You own this layer. It should follow you from job to job. Your employer has no claim on it and, in most cases, no reason to see it.
Layer 2: Work information
This layer holds the memory of the job. Projects, the decisions you made and why, the stakeholders, the current state of a launch, the documents that matter, the customer who escalated last week.
Same hypothetical PM: her assistant knows the Q3 pricing decision, the three open risks on the mobile release, which engineering lead pushes back on scope, and where the latest PRD lives.
The company owns this layer. It belongs to the role. When she leaves, the next person in the seat needs it, and she shouldn't carry it to a competitor.
Layer 3: How you work
This layer holds prompts, skills, checklists, and reusable patterns. It splits in two.
Some of it is the job's procedure. Your team's format for a launch brief. The review checklist for a pricing change. The prompt that turns call notes into a CRM update in the shape sales ops expects. These are SOPs, and they belong to the role.
The rest is personal habit. The way you ask the AI to push back on your drafts. Your morning planning prompt. Your own writing rules.
Back to the PM. She has a skill that turns customer interview transcripts into the team's standard insight template. She also has a prompt that flags hedging in her writing. The first is a team asset. The second is hers.
What goes wrong when the layers blur
I see three failure modes.
- Personal facts leak into work. If one memory store holds everything, the AI drafting a note to a client knows about your medical appointment or your divorce, and it may mention either. A shared workspace may surface those details to colleagues. Even without a leak, people who suspect one will stop telling the tool useful things, and the tool gets worse for everyone.
- Work context walks out the door. If project memory lives in someone's personal account, it leaves when they leave. The successor starts from zero, and the company's decision history now sits in a personal chat app the company can't audit or revoke. That creates a knowledge gap and a security gap at the same time.
- Good methods stay trapped. Picture an analyst who builds a prompt that cuts a weekly report from two hours to twenty minutes. It lives in her chat history. Nobody else knows it exists. When she changes roles, the team loses it.
The opposite mistake shows up too. A company scrapes everyone's prompts into a shared library, and one person's quirks get published as the standard.
How to manage each layer
Personal: the individual owns it and keeps it private
- Keep it in a store the person controls: a personal account, or a profile section the employer can't read.
- Share only the slice work tools need, such as name, formatting preferences, working hours, and accessibility needs. Let the person pick that slice.
- When someone changes jobs, this layer goes with them. Nobody has anything to hand over.
If you lead a team, don't build systems that require people to feed their personal lives into company memory to get good results.
Work: the company owns it and attaches it to the role
- Store it in company systems: project workspaces, shared docs, and team-level AI memory scoped to a project. Keep it out of personal chat accounts.
- Organize it by project or role instead of by person. Ask "where does the launch context live?" and you should get a place as the answer, with an owner's name attached.
- Apply the same access controls you use for the underlying documents.
- When someone leaves, the context stays and the successor inherits it. Write that step into offboarding next to transferring file ownership.
- Expire it. Projects end, and stale context produces worse answers. Give each workspace a review date.
How you work: sort SOPs from habits
This layer needs a promotion path.
- Start personal. People experiment in their own space. Most prompts never earn a wider audience, and that's fine.
- Promote what encodes the team's way of working. When a prompt captures how the team does a job, or a second person starts using it, move it into a shared library. Give it an owner, a version, and two lines on when to use it.
- Maintain it like any SOP. A team skill without an owner drifts out of date. Review the library on a schedule, and retire what nobody uses.
- Leave habits alone. Don't require people to publish their prompts. Ask them to share what they think others could use, and give them credit when they do.
- Check at exit. Team skills already live in the library, so they stay. Personal habits leave with the person. If a useful procedure still lives in someone's private history on their last day, offboarding should catch it.
The three layers side by side
| Personal | Work | How you work: SOPs | How you work: habits | |
|---|---|---|---|---|
| Owner | Individual | Company | Team, with a named maintainer | Individual |
| Lifespan | A career | A project or role | As long as the process exists | As long as the person finds it useful |
| Portability | Goes with the person | Stays with the role | Stays with the team | Goes with the person |
| Sharing | Private; person picks what work tools see | Same access as the project's documents | Shared library, open to the team | Private unless the person chooses to share |
What to do this week
Pick one AI tool your team uses and check where each layer lives. If personal facts, project memory, and team procedures share one store, separate them. Move personal details to each person's private profile. Move project memory into a company workspace with an owner and a review date. Move team procedures into a shared library with a named maintainer.
Then add one line to your offboarding checklist: "Hand over AI project context and promote reusable skills to the team library." That single line keeps the next departure from costing you the project's memory.